The search starts with a scope built around channels:

• Who can run paid search?
• Who can run paid social?
• Who has the cleanest account management process?

Finalists get judged on their tactical polish and platform-reported ROAS numbers in their case studies, and the team with the most confident answer to “how will you manage our accounts” usually wins. Every step of that selects for channel competence, which is the part of the job of a media buyers which platforms have spent 10+ years automating.

The thing actually worth paying for sits a level above an agency’s ability to pull channel levers, and it rarely comes up until the contract’s signed and a year has passed without the business moving.

What you actually get when you hire an agency to run a channel

Hire an agency to manage paid search and you get someone who manages paid search well. They’ll structure the account, write the headlines, tune the bids, and send a monthly report with strong ROAS on branded terms and a healthy platform-reported cost per lead. Every figure will look defensible. Almost none of it tells you whether the channel produced revenue you’d have missed without it.

Platform-reported metrics show you which channel got credit for a conversion. They say nothing about whether that conversion would have happened anyway. Branded search is the clearest case. It posts the highest return in nearly every account because every sale that touches a branded keyword gets counted as its win. But the person typing your company name into Google was usually already going to buy. The channel collects demand that other work created and reports it as its own.

An agency hired to run that channel will run it well. They’ll drive the cost per conversion down quarter after quarter and show you the trend. They’re doing the job you defined. The job you defined just doesn’t have much to do with whether the company is growing.

The one question to build your agency evaluation around: Did marketing drive net-new sales?

A marketing investment has to answer one question: did our marketing drive net-new sales we wouldn’t have gotten anyway? Impressions, clicks, leads, and platform-reported ROAS are proxies that may or may not point at that answer.

A partner who works this way starts with your economics. They ask how you define a profitable customer, what your close rates look like, and which conversions in your current reporting are real incremental wins versus demand you’d have captured for free. They treat every channel as a way to produce incremental revenue and stay loyal to none of them. When the data shows a channel taking credit it didn’t earn, they tell you, and they recommend cutting it, even when it’s the channel they were hired to run.

We saw this with a client running heavy paid brand search campaigns with a reported ROAS that made the spend look untouchable. We thought the budget was harvesting demand rather than creating it. We cut 42% of their brand search spend through a matched market geo-lift test. Organic purchases rose 61% in the test markets. An agency hired to manage paid search would never have proposed cutting brand search. As a partner hired to grow the business, we had no choice.

3 questions that separate measurement partners from channel operators

The questions you ask during evaluation decide what kind of firm you end up with. Most evaluations test for channel fluency. Instead, you should test for marketing measurement expertise. A few questions separate partners from operators quickly:

1. How do you report on marketing’s impact on business performance?
2. What KPIs do you use to make budget allocation decisions across platforms?
3. What is your approach to optimizing campaign performance based on revenue?

Operators steer back to tactics like account structure, audience strategy, and creative cadence. That work matters, but the platforms increasingly handle it. The firms worth hiring talk about setting KPIs based on business unit economics, measurement design, incrementality testing, and feeding qualified lead and revenue data back into the platforms. They’ll tell you plainly that some of your current spend probably isn’t working, and they’ll explain how they’d prove it.

What an agency’s pitch tells you about how it will work

The opening of a pitch tells you what a firm thinks it’s selling. When the slide after the introduction is a wall of platform-reported ROAS and cost-per-lead trends, you’re being sold channel reporting. When the team opens by asking what a customer is worth to you and how you currently know whether marketing is working, you’re talking to a firm built around the right question.

How a marketing agency handles uncertainty is the other signal. Operators promise to improve the metrics they control. Real measurement partners will warn you that the first honest incrementality test usually delivers an uncomfortable result, and that finding those results before your CFO does is the whole point of the engagement. That candor is the value, and it’s worth more than a flawlessly tuned account aimed at the wrong outcome.

You’re buying an answer

The agencies worth their fee from here on are the ones who can stand in front of your finance team and show, with evidence, that the marketing produced sales the company wouldn’t have made otherwise.

That’s what your evaluation should be built to find. Judge every firm in the room against the one question that decides whether the budget is worth spending at all, and you’ll stop selecting people to run channels and start selecting people to grow the business.

The post Stop Hiring Agencies to Run Channels. Hire Them to Answer One Question. appeared first on Silverback Strategies.

Where the output model came from

Deliverable counts won because they’re measurable. A retainer that promises eight posts and four service pages is simple to write, simple to track, and simple to report against. The agency hits the number, the dashboard shows the number, and the relationship feels accountable.

The trouble is that the number has almost nothing to do with whether the brand is gaining ground. A team can publish on schedule every month for a year and end up with a library of thin, keyword-targeted pages that no model ever cites and no buyer ever remembers. The output was real. The influence never materialized. Funding throughput feels responsible while quietly buying nothing that lasts.

What AI search actually rewards

When someone asks ChatGPT, Gemini, or Claude who the best provider is or how to solve a problem, the model answers with the brands and sources it has learned to trust. Recency and publishing cadence factor in. But that trust is also built from signals that accumulate across the open web: consistent mentions in credible publications, a clear and stable association between your brand and your area of expertise, third-party validation, original data that other people cite, and structured information a crawler can actually parse.

In terms of the structure, it’s often as simple as what you include within the first third of your content. A study from Growth Memo found that 44% of AI citations come from the intro. LLMs look for the most important things early on in your content. As a result, frontloading your structure with your strongest data can have a meaningful impact.

Those signals come from being referenced, recommended, and treated as a source by the parts of the internet the models already trust. Publishing volume doesn’t necessarily equate to trust. The work that builds trust looks different from a content calendar, and it shows up in your SEO budget differently too.

Authority compounds. Output depreciates.

A thin blog post written to rank for a keyword starts losing value the day it goes live. The next algorithm shift, the next competitor, the next change in how AI surfaces answers can erase it. You’re renting a position you have to keep re-buying.

Authority works the other way. A piece of original research that gets cited keeps earning citations long after it’s published. A clear entity association strengthens every time another credible source reinforces it. A reputation as the brand that actually understands a category spreads and gets harder for a competitor to dislodge. Volume keeps you on a treadmill you pay to stay on. Authority is an asset that keeps paying after the invoice clears.

How to budget for influence instead

Rebuilding the SEO line starts with scoping it around the authority you’re trying to build rather than the deliverables you’re trying to count. A few shifts do most of the work.

Fund the assets the rest of the internet wants to reference. Original research, proprietary data, and genuinely useful tools earn the links, mentions, and citations that volume content never will. Fund third-party presence through digital PR, expert commentary, and real participation in the communities and publications your buyers and the models both read. Concentrate spend on a small number of cornerstone assets that define your expertise, built deeply, instead of spreading it thin across many shallow pages. And fund the technical and structured-data work that makes your authority legible to crawlers and AI systems, so the trust you’ve earned is something the models can actually see.

Then measure the thing you’re now paying for. Published content count and keyword rankings describe the old model. Influence shows up as share of voice in AI answers, citations and brand mentions across credible sources, and growth in demand for your brand and category.

What to tell your CFO

The output model is easy to put in a contract and hard to defend once someone asks what business results it produced. Funding influence changes that conversation for the better. You’re paying for durable visibility in the channel where more of your buyers now start their research, and that visibility behaves like an asset that compounds rather than a cost you repeat every month. Tie the spend to whether the brand is showing up and getting cited where buyers are looking, and you have a line you can actually defend.

The brands winning in AI search are the ones the models have learned to trust. Authority is what earns that trust, and authority is what your SEO budget should be buying.

The post Stop Funding SEO for Output. Fund It for Influence. appeared first on Silverback Strategies.

Ad platforms optimize toward the last thing it can see

Every bidding algorithm needs a conversion to optimize against. When the only signal you send back is a form submission, the platform treats that submission as the goal and spends your budget finding people most likely to submit one. Those people aren’t always the ones most likely to buy. Plenty of them are easy to convert on a low-commitment action and hard to move toward a real purchase.

The result looks fine in the dashboard. Cost per lead drops, lead volume climbs, platform-reported ROAS holds up. Meanwhile, the sales team says the leads are junk, the close rate is sliding, and the real cost to book a customer is rising even as every number the platform reports keeps improving. The reporting and the reality have come apart.

Why do industries with long sales cycles struggle most with media measurement?

Offline conversion tracking matters most when the sale closes away from the website, and that’s the norm in home services, higher education, and healthcare. Long sales cycles and offline closes are the rule in all three, which is why the gap does the most damage in these categories.

In home services, the lead is a request for an estimate. The revenue is the booked job, often days or weeks later, after a sales visit and a quote. In higher education, an RFI inquiry sits at the top of a nurture cycle that can run months or years before an application, an acceptance, and an enrollment. In healthcare, the call or form is the start of a path through scheduling, a consult, and sometimes a procedure, all governed by compliance rules that limit what data can be passed between platforms. In all three, the moment that matters to the business sits far downstream of the moment the ad platform records and optimizes against.

Offline conversion tracking starts with the signals you already have

Closing an offline conversion gap means sending real outcomes from your CRM back to the ad platform instead of letting it optimize on raw form fills. The data usually already exists in your CRM. You must connect it back to the ad platform – either directly or through value-based proxies – so the algorithm learns from booked revenue and outcomes that are directional to revenue, rather than from raw lead volume.

Value-based bidding needs a signal that’s fast, frequent, and predictive of revenue.

Start by defining the conversion that actually matters to the business: a booked job, an enrolled student, an attended appointment. That’s the number you run the business against, and it belongs in your reporting.

It’s usually the wrong thing to bid on, though. Bidding algorithms learn from signals they see often and recently. An enrollment that lands nine months after the first inquiry, or a job booked five weeks after the estimate, arrives long after the auction it should have influenced. Feed the platform something that slow and there’s nothing there for it to learn from.

So do the legwork to find the middle event: the earliest thing after the lead that reliably separates good leads from bad, ideally within about a week. In higher ed that might be an application start or a completed advisor call. In home services, a set appointment or an issued quote. In healthcare, a scheduled consult. Pull a few months of CRM data, group leads by which early event they hit, and compare close rates. The signal you want does two things: leads that hit it close at a much higher rate than leads that don’t, and it fires often enough to give the platform steady weekly volume.

Then pass it back. Google’s enhanced conversions and Data Manager API, offline conversion import, and the equivalent tools on Meta let you send these outcomes back in a privacy-safe hashed format. Optimize toward the early predictive signal, and keep sending the true downstream outcome as well, for reporting and to confirm the proxy still predicts what you think it predicts. That relationship drifts as the offer, the market, and the sales team change, so check it a couple of times a year.

Value still matters. Give each early event a value based on what it’s historically worth downstream. If a set appointment closes 40% of the time at an average job size of $12,000, it’s worth roughly $4,800.

Healthcare needs an extra layer of care. You can still close the loop using hashed, de-identified, and modeled conversions that keep protected health information out of the platforms. Done right, it sends back enough signal for the algorithm to stop chasing appointment-request forms that never become patients, without ever handing a healthcare CRM to Google.

What changes when an ad platform sees real revenue indicators?

Once the platform optimizes against booked outcomes, it starts finding the people who close rather than the people who fill out forms. Spend moves toward the campaigns, audiences, and creative that produce customers and away from the ones that produce cheap leads or wasteful bot traffic. Cost per lead usually rises, which tends to worry people until they notice the cost to acquire a real customer dropped (the number that mattered in the first place).

It also gives you an honest way to compare channels. A channel producing expensive leads that close at a high rate can easily beat one producing cheap leads that almost never close, and that difference stays invisible until close-rate and revenue data flow back into the system. When Silverback made this shift with a healthcare client, qualified leads increased 2x.

Stop optimizing toward what you can see

The platforms have gotten very good at hitting whatever target you hand them. For long-cycle, offline-close businesses, that’s the core risk, because the easy target to hand them is the form fill, and the form fill isn’t the sale. Send the algorithm the outcome you actually care about and it will chase that instead. Until you do, you’re paying a sophisticated optimization engine to get better and better at the wrong job.

The post Offline Conversion Tracking for Long Sales Cycles: Why Home Services, Higher Ed, and Healthcare Marketers Are Flying Blind appeared first on Silverback Strategies.

Most news coverage is treating this as a loss of control. For the average advertiser, it’s closer to an upgrade, because the control you’re losing was control over a network that was wasting a chunk of your budget anyway.

What’s actually changing

The Google Display Network (GDN) inventory isn’t going anywhere. What’s going away is the workflow where you build a Display campaign and hand-pick the placements it runs on. With Demand Gen campaigns, Google leans on Gemini to find the people most likely to act and serves them across the same network, alongside YouTube, Discover, Gmail, and now Maps. Collapsing Google Display Network into Demand Gen shifts the orientation from where the ad runs, to who sees it.

Your access to the network is intact. Only the campaign shell around it changed. Placement exclusions still work, and channel-level reporting gives you cleaner visibility into how the Google Display Network performs against the other surfaces than standalone Display ever did. You also pick up tools the Google Display Network never had: lookalike segments for prospecting, carousel and expanded video formats, and generative creative. If you want to stay GDN-only, channel controls inside Demand Gen let you do that (but we wouldn’t recommend it).

Why losing display ad placement control is mostly a gift

Hand-picking Display placements always felt like control. In practice it was a part-time job spent fighting a network that defaulted to junk. Standalone Display was notorious for made-for-advertising sites, cheap clicks that never converted, and bot traffic that inflated every metric except revenue. You could prune placement lists every week and still bleed budget into inventory that was never going to produce a customer.

Audience-led targeting changes the job for marketers. Instead of curating where the ad shows up and hoping the right audience follows, you point the system at the people likely to convert and let it find them across the network. For most accounts that beats manual placement curation, because the algorithm now optimizes toward the outcome while you were previously optimizing toward a proxy for it.

The performance case for leaning into Demand Gen campaigns

In Silverback’s incrementality tests, Demand Gen has delivered as much as 6x ROAS, often 2-3x more ROAS than Google attributes to Demand Gen campaigns in-platform.

A channel that looks mediocre on platform-reported numbers can be one of the strongest performers in the account once it’s measured against incrementality. Plenty of advertisers were ready to cut Demand Gen on its reported ROAS, the same way they would have cut a weak-looking standalone Display report. Folding Google Display Network into Demand Gen pushes more budget toward a campaign type that tends to be undervalued by the reporting most teams still rely on. If you’ve built any kind of incrementality testing into your measurement, this migration works in your favor.

The catch: Demand Gen campaigns only pay off if you stay involved

Demand Gen rewards advertisers who keep doing the unglamorous optimization and signal cleanup work and punishes the ones who treat it as set-and-forget. To be successful you must:

• Keep your placement exclusion lists clean and current, because brand-safety and junk-placement risk didn’t vanish when the campaign type changed to Demand Gen.
• Watch the channel-level placement reports for where spend is actually landing.
• Monitor for bot and low-quality traffic the same way you always should have on the Google Display Network.
• Resist the auto-upgrade button on day one. 

Our recommendation for navigating the Google Display Network (GDN) to Demand Gen migration: Build a fresh Demand Gen campaign with channel controls set to GDN, run it alongside your existing Display campaign, and compare them on incremental outcomes before you commit. While you’re at it, audit your creative, because Demand Gen’s feed and video surfaces expect high-resolution imagery, multiple aspect ratios, and short-form video. Static banners alone will hold you back.

Take the upgrade

The migration is happening on Google’s timeline whether or not you like the new interface. Use the transition period as an opportunity to familiarize yourself with the power (and limitations) of Demand Gen for your brand.

In reality, Google is retiring a campaign type that most advertisers had already outgrown, and your budget is better off for it.

Want some fresh eyes on your campaigns as you make the switch to Demand Gen campaigns? Contact us and our team can provide a tailored roadmap for your business.

The post Google Is Killing Display. That’s Good News for Your Budget. appeared first on Silverback Strategies.

The fix isn’t abandoning content production, nor is it producing less of it. It’s setting a different target for each of the three search types, informational, commercial, and transactional, and building an SEO and AI search strategy around that distinction instead of writing the same “rank for the keyword” post regardless of what the person searching actually intends to do next.

Informational queries: the target is the mention, not the click.

The mechanism behind this shift is worth understanding before reacting to it. 

Roughly 99.2% of keywords that trigger AI Overviews carry informational intent, which is the exact category that “how-to” guides, definitional explainers, and general reference posts have always occupied. That concentration is a direct extension of how AI Overviews and generative search have reshaped search behavior over the past year, and it explains why this category is absorbing nearly all of the visible traffic decline while other categories remain comparatively stable.

The appropriate response is to change what the content is built to do. Content written to answer a simple question no longer earns a click for answering it, because a model can synthesize that answer directly on the results page. What it can still earn is a citation or a mention in AI results.

Original research, named data points, benchmark studies, and canonical explainers grounded in real audience research rather than generalized assumptions about what a topic’s audience wants to know help a brand differentiate themselves for inclusion in AI recommendations. Brands cited in an AI Overview still see approximately 35% more organic clicks than brands without a citation. AI-referred traffic also converts up to 23x higher than traditional search traffic. Capturing the citation and being mentioned in the answer are meaningful outcomes to optimize for in this category, rather than the ranking position.

Commercial queries: the target is the decision, not the definition.

Commercial, comparison-style searches occupy a genuine middle ground, and they deserve a different treatment than either of the other two categories. AI Overviews are expanding quickly here, up 71% over a recent six-month period, and comparison-format queries specifically trigger them in over 95% of cases, a trend consistent with the broader shift in where Google and OpenAI actually stand heading into 2026. At the same time, the intent behind these searches is different from a purely informational one. This is the search behavior that defines most B2C purchase research, where consumers weigh several options before committing, particularly for longer research cycles like finance, travel, or software, and generally still want to click through and verify claims directly rather than accept a summarized answer at face value.

The content that performs best in this category isn’t a generic “top 10” list that a model can compress into two sentences without losing anything of value. It’s comparison content grounded in direct testing, firsthand experience, or proprietary data, detailed enough that it becomes source material for the AI answer while still giving a research-stage buyer a reason to click through and confirm the details themselves. This is the one category where writing to be cited and writing to be clicked aren’t competing goals, and a single well-built asset can accomplish both simultaneously.

Transactional and navigational queries: the target is still the click.

This category is the clearest exception to the broader trend, and it’s worth stating plainly rather than treating it as an afterthought. AI Overviews barely touch transactional intent, appearing only 5 to 8% of the time overall, and e-commerce queries specifically sit closer to 3 to 4%. The reason is straightforward: a summarized answer cannot complete a purchase, book an appointment, or add an item to a cart, so someone ready to take that action still has to land on a page to follow through, which is exactly the behavior ecommerce SEO programs are built to capture.

Product pages, category pages, service pages, and local pages should continue to be optimized for conversion using the same fundamentals that have applied for years, an area where website optimization and conversion-focused design still do most of the work. This is not the part of the content strategy that needs to be reconsidered or deprioritized in response to AI Overviews, and treating it as though it does risks pulling attention away from the pages still generating the most direct commercial value.

Informational content isn’t ending because content stopped mattering. It’s ending because clicking stopped being the only reward for that one particular type of content. What that means in practice is that a content strategy now has to run three distinct targets simultaneously rather than applying one blanket approach across the entire library: mention-worthy work for informational searches, decision-worthy work for commercial searches, and conversion-worthy work for transactional ones. 

The brands adapting well to this shift are producing the right kind of content for each type of search, rather than continuing to treat every search as though it were the same request for the same outcome.

The post Stop Writing One Kind of Content for Three Types of Searches appeared first on Silverback Strategies.

Traditional SEO tactics are still driving real B2C revenue.

Proven SEO tactics still convert well, because most consumer purchase journeys still start with a traditional search. When you build category pages around search intent, product pages that are structured to rank, maintain a technical foundation that drives silent growth, and backlinks from sites that matter. These tried-and-true tactics are still as relevant as ever, just with an AI layering a new level of complexity.

According to data from Opensend, organic search still accounts for roughly 53% of ecommerce traffic and drives about 31% of ecommerce revenue, making it the largest channel most consumer brands have. And marketers aren’t quietly walking away from it. A full 92% say they’re maintaining or increasing SEO investment in 2026, not cutting it in favor of something newer.

Most “SEO stopped working” conversations are actually “we never fixed the technical foundation“, ignoring accrued technical debt or lacking core content needs. A product page ranking #3 for a high-intent term and a product page that doesn’t rank at all are not the same starting point, and no AI-visibility layer fixes a page that was never findable to begin with.

Brand mentions in AI overviews are a trust signal 

AI Overviews and chat platforms are a new visibility view point stacked on top of search, not a replacement channel that makes the old one obsolete. Treat it that way and you’ll spend a year optimizing for the wrong thing.

AI Overviews now show up on roughly 13% of U.S. desktop queries, double the rate from January 2025. That’s a real shift, and it’s exactly what we broke down in how AI Overviews and generative AI are reshaping search behavior. But here’s the part that gets missed: brands cited inside AI Overviews see 35% higher organic click-through and 91% higher paid click-through than competitors on the same query who aren’t cited. Visibility inside AI answers compounds the traffic you already have. It doesn’t cannibalize it.

Google still sends roughly 345 times more traffic than ChatGPT, Gemini, and Perplexity combined. AI is an additive share right now, not a majority share of how users are finding brands. Build your AI visibility strategy on top of a winning SEO strategy for the best results.

AI can’t see far enough into your business to replace human judgment.

Here’s the limitation nobody wants to say out loud: AI doesn’t know your actual customer, your current inventory, your return policy exceptions, or what changed about your brand last week. It’s working from patterns seen across the web, not specifically what you need at the moment for the business.

Consumers can feel the difference. 73% say they trust AI-generated content in general, but 52% reduce engagement the moment they identify a piece of content as AI-written. That’s not a knowledge gap, it’s a trust gap, and it’s specifically about authenticity and overuse of AI-written content. The share of consumers who say heavy AI use would decrease their trust in a favorite brand doubled from 20% in 2025 to 40% in 2026. Half of U.S. consumers now say they’d rather give their business to a brand that doesn’t use generative AI in customer-facing messaging at all.

This is exactly where real audience research and human-written or human-edited content earn their keep. AI can help you scale and structure content, but cannot do the qualitative work of knowing why your specific customer buys, what objection is actually stopping them at checkout, or what your brand sounds like when it’s telling the truth instead of a templated version of the truth.

The combined playbook is what actually works for B2C brands right now.

The strategy is simple but takes real discipline to run: proven SEO foundation, a structured AI-visibility layer on top of it, and human research and editorial judgment sitting over both. Human-edited AI drafts are consistently outperforming both fully-AI and fully-human content on organic click-through and ranking metrics this year. When leveraging these together, you gain the benefits of both systems without the drawbacks of either.

None of this means the question was wrong to ask. It means the honest answer is more specific than ‘yes’ or ‘no.’ Start by auditing your technical and content foundation before you reallocate budget toward whatever’s trending this quarter, then layer in structured, citable content built for both people and AI on top of it, tracking AI citations alongside your rankings instead of in place of them. Keep a human review gate on anything customer-facing, because consumers can tell when that’s missing and it changes whether they trust you. Maximise audience research as it’s a key tool in 2026 that can make the difference between getting lots of visibility and earning revenue from high intent users.

SEO in 2026 isn’t the SEO of the past decade,but it isn’t obsolete either, the way we laid out in where Google and OpenAI actually stand heading into 2026. The risk was never that SEO stopped working. It’s brands mistaking a shift in the judging criteria for a reason to gut the program that’s still driving most of their revenue.

The post SEO Isn’t Dead, It’s Just Not Alone Anymore: The B2C Case for SEO + AI. appeared first on Silverback Strategies.

Google has spent several product cycles moving control away from the literal search term an advertiser bought and toward its own read of intent. 

This piece covers what’s actually changed in how search term matching works, why fighting to reclaim that control is a low-ROI endeavor, and where to focus your optimization efforts instead.

What Exact, Phrase, and Broad Match Keywords Actually Do Now

The three keyword match types still exist in the Google Ads interface, but none of them work the way they did when the system was built.

Exact match no longer means the literal query. It accepts close variants — different word order, added or removed function words, and queries Google judges to have the same meaning as the term you bought.

Phrase match has absorbed the behavior that used to belong to modified broad match: it captures the intent and word order of your phrase while allowing additional words before, after, or in between.

Broad match has moved the furthest from its original purpose. It no longer primarily matches on the terms in your keyword list. Instead it reads signals: your landing pages, the other keywords in your account, the searcher’s context, and your bid strategy, then decides which queries are worth showing your ad against.

The common theme? Google is leaning harder on its own inference of intent than on the exact text an advertiser purchased. Each match type moved in that direction independently, and together they add up to a meaningfully different system than the one paid search precision was originally built around.

Where Match Types Disappear Completely: Performance Max and AI Max

Two Google products take the search term matching evolution to its logical end.

Performance Max removes match types altogether. In their place, advertisers provide Search Themes — a directional input describing what a campaign is about, not a control that determines which queries an ad can or can’t match.

AI Max layers automated query expansion and AI-generated creative directly onto standard Search campaigns, rather than existing as a separate campaign type. Search term matching moves beyond the keyword list entirely, ad copy gets generated and tested automatically, and in some configurations the system can route traffic to landing pages of its own choosing. 

Google has also been shifting parts of this from an opt-in feature to the default behavior for eligible campaigns. Automation is becoming the assumed state of a Search account, not an optional layer on top of one.

This is a significant reduction in the granular control paid search teams used to have. Search Themes are not match types with a different label, and automated query expansion is not the same discipline as a curated keyword list.

Why Fighting to Reclaim Keyword Control Is A Low-ROI Endeavor

This keyword evolution doesn’t mean your keyword strategy is worthless or that match type selection doesn’t matter anymore. It does. But the return on hours spent trying to force the old level of precision back into an account keeps shrinking. 

Exhaustive negative keyword lists built to counter broad match’s signal-reading behavior, granular match type experiments designed to outmaneuver an increasingly inference-driven system, endless query report audits chasing control that the platform is actively moving away from — these activities used to be where a strong paid search team distinguished itself. 

Increasingly, they’re hours spent simply working against the direction the platform is going. Time is better spent building the inputs that actually shape how well the automation performs.

Customer Research: The Input the Algorithm Can’t Generate for You

An inference-driven system like Google Ads is only as good as its understanding of who a good customer actually is, and that understanding has to come from somewhere outside the platform. Google’s models can read landing pages and search context, but they can’t tell you why your best customers actually buy, what objections come up in a sales call, or which segment of “qualified lead” churns six months later.

This is where you can build real leverage now: analyzing which queries and audiences are actually converting into revenue, not just clicks; building a feedback loop with sales or CRM data on lead quality; and talking to actual customers about why they chose you. That research becomes the intent signal that the algorithm optimizes from.

Creative and Asset Quality Fuels Ad Algorithms

Automated targeting and automated creative both work from the same underlying material: your landing pages, ad copy, sitelinks, and other assets. A well-targeted campaign built on a thin or generic asset library still caps out quickly, because there’s little for the system to differentiate or learn from.

Practical work here includes disciplined creative testing rather than a “set it and forget it” asset library, landing page relevance work that gives automated matching something specific to read signal from, and active management of pinned or excluded content so brand-critical language stays intact even as the system generates more around it. 

This is ongoing work, and it’s a better use of a paid search team’s time than another round of match type adjustments.

First-Party Data: The Signal Automated Bidding Actually Needs

Automated bidding systems optimize toward whatever conversion data they’re given, and a thin or generic signal produces thin, generic optimization — a pattern paid search teams are increasingly running into across bidding, not just matching. 

Conversion action hygiene, accurate value reporting, and clean audience and feed data all directly shape how well an inference-driven system performs, in the same way that landing pages and creative do.

The Job of a Paid Search Manager Shifted

The value of precise keyword by keyword optimization is eroding, but the skill underneath that precision — understanding intent, building the inputs that shape performance, and checking results rigorously — hasn’t gone anywhere. 

Modern paid search managers have shifted focus to research, creative and asset quality, first-party data, and measurement.  

If your team is still spending most of its time on keyword and match type management, or you’re not sure your account is set up to feed AI Max and Performance Max well, Silverback can help you figure out where your paid search hours are actually worth spending. Get in touch with our team to talk through what that shift looks like for your account.

The post Google Loosened Keyword Control. Here’s Where Your Team’s Time Should Go Now. appeared first on Silverback Strategies.

Starting August 17th, 2026, Google is changing the bidding behavior that made that possible, and the accounts currently coasting on it are about to see their numbers move. 

Most of the coverage of this change treats it as a deadline: update your targets in the Bid Target Adjustment Tool, or watch your CPA rise. That’s true, but it’s the smaller story. The bigger one is that the gap you’ve been benefiting from wasn’t a mistake — it was room Smart Bidding used to test and explore on your behalf — and this update closes it. 

Here’s what’s actually changing, why the “overperforming” target was never free, and how to decide what your bidding strategy should do next, rather than just resetting a number to clear the deadline.

What Is Google’s August 17th Bidding Update?

On August 17th, 2026, Google is changing how target-based bid strategies behave on campaigns marked “Limited by budget.” Today, when a budget-limited campaign runs Target CPA or Target ROAS, Smart Bidding often delivers better than the target you set because it’s concentrating a capped budget on the cheapest, most efficient auctions it can find. 

After August 17th, that stops. Budget-limited campaigns will optimize consistently toward the number you actually entered, including when you adjust budgets.

Google’s own stated example makes the stakes concrete: if your Target CPA is $10 but your recent actual CPA has been running at $5, your campaign will start delivering closer to $10.

Why Is My Google Ads CPA About to Go Up?

Because the gap between your stated target and your actual performance was never a permanent feature of your account. It was a byproduct of how Smart Bidding handles a constrained budget. 

When spend is limited, the algorithm doesn’t need to spend all the way up to your target to hit your budget cap, so it skims the cheapest available conversions and leaves the rest of the auction alone.

After August 17th, that skimming behavior goes away. The algorithm optimizes toward your target as configured, not toward the cheapest outcome it can find under a spending cap. If you don’t touch anything, your real-world CPA drifts up (or your ROAS drifts down) to match the number you set, possibly months or years ago.

Which Campaign Types Are Affected by the August 17th Change?

The update applies to any campaign using a target-based bid strategy that can be marked “Limited by budget”:

• Search
• Shopping
• Performance Max
• Demand Gen (including Target CPC, which functions as the target-based strategy for this campaign type)
• Travel

Display and Hotel campaigns already operate this way, so nothing changes for them. App campaigns and Video reach/view campaigns are excluded entirely.

Worth noting for anyone running Performance Max or Demand Gen: Google has said advertisers may see traffic shift between channels within these campaign types as the system rebalances toward the target. A tighter CPA isn’t the only thing that moves; where your budget actually gets spent can shift, too.

Was Your “Overperforming” Campaign Actually a Mistake?

Most coverage of this update treats the overperforming target as an error to be corrected: set your target above where performance settled, and the discrepancy is a loose bolt to tighten before the deadline. That framing misses what was actually happening.

A target set above your true cost goal gives Smart Bidding room to work. It can bid into less certain auctions, test new audiences, and surface conversions it wouldn’t touch under a tighter constraint, and over time, some of that exploration turns into the cheaper, durable performance you’ve been benefiting from. The “overperforming” number was slack you were, knowingly or not, using to let the algorithm find things a tighter target would never let it try.

August 17th closes that room. Any account currently coasting on a gap between target and actual performance should assume the exploration budget that produced it is going away.

Why Google’s Answer Is “Raise Your Budget” And Why That Deserves Scrutiny

Google’s framing for this update is predictability: campaigns will perform more consistently at your stated target, including when you increase budget, which makes scaling easier to plan around. 

That’s a real benefit, and it’s true on its own terms.

It’s also an answer that happens to serve Google. Once a budget-limited campaign can no longer significantly beat its target, the most direct lever left for growing volume is raising the budget — spending more to get more, at a target you now have to actively manage. 

That’s a reasonable option. It’s not the only one, and it’s worth noticing that it’s the option Google’s own tooling and messaging point you toward first.

Treat “raise your budget” as one input to a decision, not the decision itself. The right response to this update depends on what you actually want from the campaign — protect current cost efficiency, accept a higher CPA in exchange for volume, or hold and watch — not on which lever Google made easiest to pull.

How to Prepare for the August 17th Bidding Change

1. Audit every budget-limited campaign on Target CPA, Target ROAS, or Demand Gen Target CPC. Pull the gap between stated target and trailing actual CPA/ROAS for each one.
2. Decide deliberately for each campaign, rather than defaulting to one approach across the account:
   1. Hold the target and accept CPA rising to meet it, if the target reflects what you’re actually willing to pay.
   2. Lower the target to match recent performance, if you want to protect the efficiency you’ve been getting.
   3. Treat the gap as a real growth signal and raise the budget with a specific hypothesis for what that spend should buy.
3. Use the Bid Target Adjustment Tool as one input, not the final word. It shows recent performance; it doesn’t know your margins, your goals, or how much volatility you can tolerate this quarter.
4. Build in a recheck cadence. The gap between target and actual reopens as performance moves over time. Fixing it once in August doesn’t mean it should stay fixed.

Turn This Deadline Into a Strategy Decision

The deadline framing that’s dominated coverage of this change — update your targets or lose performance — isn’t wrong, but it’s incomplete. The bigger shift is that the era of banking efficiency on a capped budget, without ever having to explain why your numbers looked the way they did, is over. 

Campaigns that come out ahead after August 17th won’t be the ones that happened to have a favorable gap between target and actual performance. They’ll be the ones with a target-and-budget strategy someone actually chose and can defend.

That’s the work worth doing before the 17th: not just resetting a target to clear a deadline, but deciding, campaign by campaign, what you actually want your bidding strategy to do next.

Not sure which of your campaigns are exposed, or what the right call is for each one? Silverback works with advertisers to turn moments like this into a real bidding strategy, not just a settings update. Get in touch with our team to review your accounts.

The post Google’s August 17th Bidding Update: Why Your CPA Is About to Rise (and How to Stop It) appeared first on Silverback Strategies.

That’s changed. 

“The job of paid media managers really changed from trying to control all of the levers to more focusing on what are the signals I’m sending the platform so it can learn what’s a profitable customer for me and then what are the guardrails that I’m setting.”

It’s tempting to hear “fewer levers” and assume the job got easier. It didn’t. “Maybe fewer levers, but those levers matter more.” The platforms make far more decisions now. What’s left for a human to decide is what to tell the machine before it starts pulling them on its own. 

Skip that step, and the machine doesn’t pause to ask. It just runs.

What Happens When AI Controls Your Ad Creative Without Explicit Settings

The clearest example of what goes wrong sits in creative, an area that seems like it should be the easiest thing to hand over to AI.

“Meta has like 32 settings now that are all related to AI content,” Jordan says, “and you have to set them at the ad level.” The detail that matters most: many aren’t opt-in. They’re set automatically.

What does opting in by inaction look like? “Meta can just put a random song on your ad, and I’ve seen it happen. Rap music with a lot of profanity, for example, on an ad from a super polished brand that would never approve that.”

The platform did exactly what it was configured to do. That’s the point. The configuration is the decision, whether or not anyone consciously made it. 

Why AI-Driven Campaigns Attract Bot Traffic and What It Costs You

If creative automation is the visible risk (the one that shows up as an embarrassing ad, for example), there’s a quieter risk underneath it, in what platforms are told to optimize toward.

“This bot traffic has gotten even worse with AI,” Jordan says. Conversion types like click-to-calls and lead form fills are “really susceptible to bots,” and if a campaign optimizes toward raw lead volume without distinguishing real prospects from bot noise, the platform delivers exactly what it’s told to value. More leads. More clicks. More calls. None of it is necessarily real.

This connects to where ads actually run. Jordan flags Google Display Network, Google Search Network, and Facebook Audience Network as placements that tend to be “bot-heavy,” places where monitoring and explicit settings matter more than a top-line number suggests.

The throughline: AI systems are extremely good at optimizing toward whatever signal they’re given. They can’t ask whether that signal is the right one. That’s still a human job.

How Machine Learning Actually Learns (and What It Needs From You)

There’s a second layer here, and it’s counterintuitive for media managers trained in the “more control, more segmentation” era.

Since iOS tracking changes limited how precisely platforms can target individuals, Meta leans harder on pattern recognition based on behavior and creative response. 

Jordan’s example: “Jordan is active on Reels at 9pm, and spends a lot of time watching videos about dogs. Jordan behaves very similarly to these thousands of other people in the U.S.” The platform builds a behavioral fingerprint and matches ads to people with similar patterns. So,  creative becomes the primary lever, since granular audience targeting is largely gone.

But this kind of pattern recognition needs volume. “For Meta to be good at that, it has to have enough data to build a good pattern.” 

The thresholds: Meta needs roughly 50 conversions a week at the ad set level to exit “learning mode” and perform consistently. Google’s threshold is lower, but the same dynamic exists: roughly 30 conversions over 30 days.

This is where the old instinct toward control can become a liability. A media manager trained to segment heavily — separate campaigns for separate audiences, separate ad sets for separate offers — can end up with ad sets that never hit enough conversions to exit learning mode. 

Each segment might make perfect strategic sense alone. But if no segment can hit 50 conversions a week, none of them get the chance to perform the way the algorithm is designed to perform once it’s learned.

The job now: “finding the middle ground between having enough control over your budget and where you want to spend, but also giving the platform enough data to learn from.” 

Too much segmentation starves the system. Too little structure hands over decisions that should stay with a human. The guardrails sit in that middle ground.

The Paid Media AI Guardrail Checklist: Five Settings Most Accounts Get Wrong

A consistent set of things to check because each is so often left on a default nobody chose deliberately.

Brand Exclusions

Performance Max, left at default, “will automatically serve both branded and non-branded terms,” and because branded terms tend to be the cheapest to win, PMax gravitates toward them fast.

Roughly 90% of accounts audited don’t have this addressed, meaning a campaign meant to reach new prospects often functions as a branded search campaign in disguise.

Platform Optimization Targets

An estimated 80-90% of accounts optimize toward leads and phone calls without feeding back which converted further down the funnel, or which were spam. 

Without that feedback, the platform can’t tell a real prospect from a bot and has no reason to try.

Where Ads Run

Display Network, Search Network, and Audience Network placements carry higher bot exposure. These placements need active monitoring and explicit settings. 

Left on default, the platform decides where to spend.

Tracking Infrastructure

With up to 50% of Americans using ad blockers that block pixel-based tracking, server-side tracking — a direct connection between the ad platform and a business’s server, bypassing the browser — recovers data that would otherwise be lost. 

Without it, the platform decides based on an increasingly incomplete picture.

Campaign Structure & Learning Thresholds

Instead of maximizing segmentation, build structure around whether each ad set can realistically hit the conversion volume (50/week for Meta, 30/30 days for Google) needed to exit learning mode and perform as designed.

None of these are complicated on their own. What makes them matter: almost none of them are decisions a platform will surface or flag. 

They’re settings, and settings, left untouched, are still choices. Nobody made them on purpose.

What Good Paid Media Management Looks Like in an AI-First World

If the work shifted from pulling levers to setting guardrails and reading signals, what makes someone good at the job shifted, too.

“Previously a successful media manager used to over-index on detail-orientation, because a lot of the job was about control,” Jordan says. “Now we’re really screening for big-picture thinking, connecting what’s happening in a campaign to what it means for the business.”

Client communication matters more, not less, as automation increases. “At the end of the day, businesses are going to need someone to hold accountable, someone to make decisions, and communicate those decisions well.” As more execution becomes invisible, someone still has to explain what’s happening and why.

Deductive reasoning and structured diagnosis matter too: “These AI tools are probably going to spit out thousands of suggestions, but knowing what to do with that is going to be a big one.” The volume of output isn’t the bottleneck anymore. Judgment about what matters is.

And creative strategy and customer research (understanding who the audience is and how to reach them) is “more and more an important job of the media manager,” precisely because creative is now the primary lever left for influencing how these systems perform. The data backs this up — see what $1.3 billion in Meta ad spend reveals about creative performance.

The Bottom Line on AI-Driven Paid Media

This isn’t an argument against AI in paid media. These systems perform well when given what they need. The argument is narrower, and more urgent: these systems need to be given something. Signals. Exclusions. Volume thresholds. Guardrails. 

None of that happens automatically, no matter how automated the rest of the campaign looks.

“Automated” describes what happens once the inputs are set. It says nothing about whether those inputs were ever chosen on purpose. The businesses getting hurt by AI-driven paid media right now usually aren’t the ones using AI badly. 

They’re the ones who never realized there was a decision to make in the first place.

The post Higher Ad Costs, Slower Growth, Wrong Leads: What Happens When You Let Paid Media AI Run Without Guardrails appeared first on Silverback Strategies.

On one side: the marketing reports. Cost per lead, flat or improving. ROAS, climbing. By every metric the marketing team’s been trained to report on, things are working. 

On the other side: the P&L. Revenue, flat or sliding.

“Our business is either flatlining or maybe even declining a little bit, whereas our board expects us to be growing at a 15% year-over-year clip,” Jordan says, describing a version of this conversation she’s had more times than she can count. “And I, as the client, am frustrated and also confused because look at the data the ad platforms are saying. It says CPL is either flat or even getting a little better. ROAS is up.”

This isn’t a story about bad data. The platform numbers are accurate, in a narrow sense. The problem is what those numbers are measuring and what they’re not.

Why Summarizing Data Can Make Real Problems Disappear

It’s not that anyone’s avoiding the data. The team in this scenario is looking at their numbers. Regularly, carefully, in good faith. The problem is what those numbers are doing before anyone sees them.

Every report is a rollup. Daily numbers roll into weekly, weekly into monthly, channel-level into account-level, account-level into the topline summary that lands on an executive’s desk. 

Each layer of rollup is a layer of averaging, and averaging is exactly the operation that can make a real problem disappear.

A decline in one specific, important slice of the business can be small enough, relative to the whole, that it simply doesn’t move the aggregate number. The aggregate looks fine. The report says things are fine. And the slice that’s actually in trouble never gets its own line because nobody asked for one.

This is different from the goal-setting and attribution issues that create the initial gap between platform metrics and business reality. This is what happens after that gap exists: once a misleading number is baked into the reporting, the normal process of summarizing data for an executive audience doesn’t surface it. 

If anything, summarizing makes it less visible because summarizing is, by design, the process of making smaller things invisible inside bigger things.

The result is a report that isn’t lying, exactly. It’s just operating at a level of resolution where the thing that matters most doesn’t show up.

An Example: How Healthy Aggregate Numbers Can Hide a Declining Business

Let’s dive into a real client example.

Organic search performance — in aggregate — looked steady, even improving. Most teams would read that as reassuring. But aggregate organic is a blend of branded and non-branded traffic. And here, the company had built up non-branded organic visibility.

That growth was masking the fact that organic branded search clicks (people searching for the company by name, no ad involved) were “rapidly declining.” 

Nobody was looking at that number in isolation, because the aggregate looked healthy.

The business had a strong brand and great reviews. People were searching for it by name. But paid spend kept growing in branded search and Performance Max, so increasingly, the business was paying for those searches instead of capturing them for free. 

The decline in free organic brand visibility was the signal, hidden inside a number that looked fine.

Why the Warning Sign Came From Outside the Marketing Function

The most telling detail in this story is who raised the alarm.

By the time Jordan’s team got involved, the business had already let go of its in-house marketer. For six months, the reports told a consistent story: paid media performing well, path to growth is more budget. The business followed that recommendation, but revenue didn’t follow the increase in spend.

It was the company’s private equity firm — looking at the business from the outside, focused on the P&L — that called it. And, Jordan says, “They were right that that budget wasn’t being spent in the right way.”

This is the core of why this matters: the gap between “marketing looks profitable” and “revenue disagrees” can be completely invisible from inside the marketing function. 

Every number that function watches can look fine at the same time the business is struggling.

How to Test a Budget Hypothesis Without Risking the Whole Budget

Jordan’s team had a clear hypothesis: too much budget flowing to channels capturing existing demand instead of creating new demand. But moving a large share of a client’s budget on a theory carries real risk if you’re wrong.

So before changing anything, they tested it. Split the business’s markets into two groups. In one, Performance Max spend stayed put. In the other, it was turned down. Nothing else in the marketing mix changed during the window. Then they watched revenue in each group, focused on measuring the real-world outcome.

The result confirmed the hypothesis on both fronts. Revenue in the reduced-spend markets didn’t decline, which meant that spend hadn’t been generating revenue that wouldn’t have come in anyway. 

And organic brand search activity, tracked through GA4 and Google Search Console, picked back up in those same markets, just as predicted: when paid stopped capturing those branded searches, the free version of those same searches reappeared.

What Happened Next: 14% Increase in Revenue

With the test confirming the diagnosis, Jordan’s team reallocated roughly a third of the client’s budget out of Performance Max and branded search and into Meta and YouTube, channels better positioned to reach people who weren’t already looking for the business by name.

The result: media spend stayed flat. The business didn’t need to spend more to fix this. The money was already there, just pointed at the wrong thing. 

Top-line revenue increased 14% year-over-year.

Read the Full Case Study

The Question Every Marketing Budget Should Be Able to Answer

This isn’t “Performance Max is bad” or “branded search is wasted spend.” Both are too simple, and the right multiplier for a channel depends entirely on the business. Sometimes branded search really is driving incremental revenue, and defunding it would be a mistake.

The real question is underneath the whole story: is this spend creating revenue the business wouldn’t otherwise have, or is it capturing revenue that was coming in regardless? 

Platform-level reporting can’t answer that. It can only tell you whether the platform thinks it deserves credit. 

Answering the real question means looking somewhere the platform can’t see: the business’s own revenue, measured against itself, with and without the spend in question.

When marketing looks profitable and revenue disagrees, revenue is usually right. The work is figuring out why the marketing numbers didn’t see it coming, and increasingly, that work is testable. Not something a business has to hope someone notices before it’s too late.

The post Your Marketing Looks Profitable. Your Revenue Disagrees. Here’s Why. appeared first on Silverback Strategies.