AI (Artificial Intelligence) Insights from Seer Interactive

ChatGPT Ads UX Study: Findings from Real AI Search Sessions

Admin — Thu, 12 Nov 2026 19:00:00 GMT

TL;DR

We watched 161 recorded AI search sessions and logged every ad that showed up inside a ChatGPT answer. Most of the people we asked told us they hadn't seen any.

Testing llms.txt vs. AI Info Pages: GEO Quick Wins or Waste of Time?

Admin — Tue, 10 Nov 2026 17:53:22 GMT

TL;DR

I compared llms.txt and AI info pages to determine if these were easy GEO wins or tactics without data proof
llms.txt files failed to perform across several parameters: 0 for 2 on formal client tests, and almost no citation presence
AI info pages performed for 3 of 4 clients, and Seer's AI info page continues to be consistently cited a year after going live
A page bots never fetch can’t change an answer: one client’s AI info page got no traction as Markdown or as HTML, because AI agents weren’t requesting it
In terms of which is more likely to be cited in AI responses, the AI info page is your winner

Are llms.txt and AI Info Pages Worth Your Time?

When the GEO race started, everyone was guessing about what worked - even us data wranglers. Without evidence of what worked well, we settled for what worked quickly.

Two of those tactics keep landing in my inbox: llms.txt and the AI info page. They’re often lumped together because they look alike and are "for the bots." Both are also pitched as low-lift and quick wins.

But llms.txt and the AI info page do different jobs, and in our testing they had very different levels of success: the AI info page got picked up for 3 of 4 clients, llms.txt for none.

Here's how they differ, what happened when we tested them, and when I'd reach for each.

llms.txt vs. AI Info Page: What’s the Difference?

llms.txt and AI info pages both get pitched as ways for AI answer engines to find and fact-check your brand. But they serve different purposes, and bots may or may not read them.

Here are our own examples from seerinteractive.com:

llms.txt started as a community proposal for a plain-text file that points models to a site’s most useful pages. The AI info page has no spec at all. It’s just a web page written for a different reader.

And here’s a quick overview of how llms.txt and the AI info page differ:

Bots browse your site differently from how people do: they search an index, or fetch a specific URL live when a question calls for it. An AI info page fits that path, while an llms.txt file gets largely ignored unless a crawler goes looking for it.

When we tested the two tactics, the AI info page worked for Seer and 3 of the 4 clients who've tested it. Conversely, llms.txt failed to draw meaningful bot traffic, let alone steer it.

In Our Tests, AI Info Pages Outperformed llms.txt

We have a GEO Test & Learn Library that tracks every completed GEO test we've run for our clients - 70 completed GEO tests across 17 clients as of October 5, 2026.

Our findings from client tests of llms.txt and AI info pages were as follows:

0 for 2 on tests targeting llms.txt
3 for 4 on AI info page tests
Counting a head-to-head test where both launched the same day, llms.txt is 0 for 3

When we ran the tests for ourselves at Seer, we saw the same effects.

llms.txt Saw Almost No Traction Among Bots

In 2025 we shipped llms.txt for several clients. It was cheap and LinkedIn was loud about it, so we decided to test it.

Overall, our findings on llms.txt were underwhelming:

Two clients ran formal tests, and neither logged a single ChatGPT request for the file
Across nine total clients, answer agents barely asked for it and the net impact on AI citations was negligible

We even tried a different perspective, analyzing the performance of llms.txt in AI answers. Of about 9.2M AI answers across 100 client data sets, only 73 cited an llms.txt URL, mostly developer docs. Of those, 15 were Gemini inventing yourdomain.com/llms.txt while giving GEO advice.

We’re not the only ones seeing it. Ahrefs found that of roughly 38,000 sites with a valid file, 97% got no requests for it in May 2026. An SE Ranking study of about 300,000 domains found no clear effect on AI citations. And Google’s own guidance says Search ignores llms.txt.

Once the results came in, we stopped recommending llms.txt as a GEO tactic. Performance didn't move, LLMs didn’t cite the files, and the logs showed AI crawlers weren't requesting them.

AI Info Pages Performed for Us and Our Clients

Often we say that good science is both repeatable and reproducible. So we ran AI info page tests just like the llms.txt tests, and found a very different picture.

First, the bots actually go get the page. Here’s how agent requests for the AI info page compared with llms.txt on the two client sites where we can see bot logs, and on our own. robots.txt is in the table as a yardstick, since nearly every crawler checks it:

Every site in our logs with an AI info page saw far more agent requests for the page than for llms.txt, including the developer platform that runs both.

Second, what’s on the page shows up in the answers. Four clients have tested an AI info page, and AI answers picked it up in three:

Energy technology company: 21 of 57 tracked facts (37%) showed up in ChatGPT answers about a month after launch
Enterprise data-protection software company: Launched an AI info page and an llms.txt the same day; Gemini and Perplexity cited the page and never the file, though that’s an early read on about 10 answers in three weeks
Accounts receivable software company: The client’s team called it a win
Communications SaaS company: Bots never fetched the page, so it never had a chance to change an answer (more on that below)

Audience and industry framing stuck best, and proprietary product names stuck least. That's a good reminder to use the words your buyers do.

Outside of our Library, we also tracked a financial data and ratings firm. The page drew 439 requests from ChatGPT-User, 74% of all requests to it. ChatGPT-User fires when a real person’s question sends ChatGPT to read a page live. In a 10-answer sample of branded ChatGPT responses, about half picked up some of the page’s language, most of it partial matches.

No Fetch, No Result

The miss was the most useful result of the bunch.

The communications SaaS company’s Markdown (.md) version of the AI info page drew zero AI crawler requests in its final week, and only 1 of 30 tracked facts showed up in answers. The team relaunched it as HTML in January 2026 and got 0 of 30 on the same prompts.

Changing the format didn’t fix it, because bots weren’t fetching the page. A page that never gets fetched can’t change an answer, regardless of what's on it.

We still recommend HTML, because every AI info page that worked in our data was HTML, and HTML is what answer engines search for and fetch live. But this client shows that the first thing to check is the logs, not the content.

A Year In, Seer’s AI Info Page Still Owns the Who-Are-You Answer

We looked at a year of data to see whether our AI info page has continued to perform. Seer's AI info page went live in October 2025 and Wil Reynolds has tracked it in Scrunch since mid-November.

On the most direct brand question there is, "Tell me about Seer Interactive," the AI info page has been cited in about half of answers all year: 50.4% on average from late November through March and 52.5% from June through September, across 403 answers. It held steady even when citations on our other brand prompts dropped in late January.

Then we split the branded prompts in two. Plain brand prompts ask about Seer by name and nothing else, like “Tell me about Seer Interactive.” Persona prompts still name Seer, but they come in a buyer’s voice, with a role and a problem attached: “As a CMO whose paid media costs keep rising while organic results decline, I want to know how Seer Interactive builds integrated strategies that reduce paid dependency.”

The first is a who-are-you question. The second is a should-I-hire-you question, and the model pulls from a lot more of the web to answer it.

Plain brand prompts: On the six we track, the citation rate fell from 57% to 37% in a single step on January 26, then slid into the high 20s by late March.

Persona prompts: On the 41 persona-style prompts we switched to in April, citation rate fell from 19.5% to 7.8% around the start of June and has sat between roughly 5% and 13% since.

So the page holds the who-are-you answer and loses ground on should-I-hire-you. That’s the line between what an AI info page does and doesn’t do. It isn’t set and forget, either: the core answer held while the broader branded share drifted down.

Now the Bayesian caveat.

The llms.txt results are the better-controlled ones: subfolder comparisons and server logs, though each has only two to three weeks of post-launch data.

My confidence there is high. llms.txt doesn't get fetched. The AI info page wins are thinner. None had a control, most were read on a single platform, and one is an early read on about 10 answers. I'm moderately confident in the direction, and I expect to update.

Our own backlog has already voted, with 21 AI info page tests proposed or in flight and exactly one llms.txt test.

llms.txt or AI Info Page: Which Should You Use?

Stop asking "which tactic should I use?" and start asking "what problem am I trying to fix?" For most brands I talk to, the problem is that AI gets their brand information or narrative wrong. That calls for an AI info page.

Here’s a breakdown of what to focus on based on what you’re trying to accomplish:

What if a client wants llms.txt "just in case"? Fine. It's about an hour of work, and it won't hurt anything. Ship it, link your AI info page from it, and move on. Just don't count it as your GEO strategy.

What I'd Do Today

If you only do one thing based on this blog, build an AI info page. The condensed playbook:

Find the gaps. Run your branded prompts and write down what AI gets wrong or leaves out.
Write plain facts, with corrections spelled out. Short sections, dated numbers, and your buyers' words instead of your internal product names.
Publish it as HTML. A clean URL, a 200 status, no noindex, and no JavaScript needed to see the text. Link it from the footer and add it to your sitemap.
Check the logs before you judge anything. No fetches from ChatGPT-User, PerplexityBot, or similar agents means no test.
Measure language adoption on branded prompts. First look at 4 weeks, make the call at 12, then refresh it every quarter.

All of this fits into a bigger shift in how the web works these days. As we move fact-finding and research to AI tools and Answer engines, the focus goes from how to organize information for human consumption (clean UX, semantic markup, responsive design for every conceivable screen size or device type) to AI consumption (schema, .md files, etc.).

My colleagues, to their credit, have been saying so already. Alisa Scharf argues you should fix how LLMs see your brand before chasing AI rankings, and John Lovett lays out how to write those facts down in a brand canon.

If you want help figuring out where to focus your GEO efforts, Seer can build and implement a strategy for you. We can help create your brand's AI info page or llms.txt, or even work with you to build a brand canon. Chat with us to learn more.

I Built a Tool to Watch ChatGPT Search, and It Predicted My Next Question

Admin — Fri, 30 Oct 2026 17:04:41 GMT

TL;DR

I wanted to understand how ChatGPT searched, so I wrote a script to watch it and tracked 3 random shopping conversations: a carry-on suitcase, a cloud backup tool, and a Spanish app
For each topic, I asked an opening question plus two follow-ups in the same chat and saved the data as ChatGPT was building the answer
In 2 of the 3 conversations, ChatGPT's first search added a concern I hadn't raised yet but planned to ask about
ChatGPT started naming brands during its first search and cited brand sites in that first response for 2 of the 3 conversations
In the luggage conversation, ChatGPT cited one review site for 5 of the 9 sources on the first question, and the recommendation lined up with that site's ranking - then ChatGPT looked for further evidence to support that pick in subsequent answers
We can use this tool to help clients understand if their brand shows up in the first answer, and what ChatGPT goes looking for when someone pushes back on that answer

De-Mystifying the Black Box of AI Search

AI search is a black box. We still don’t know a lot about how it searches, where it pulls information from, or why it cites some sources but not others.

After Suganthan Mohanadasan's blog on ChatGPT search language made the rounds in our internal Slack, I was inspired to figure out how to reverse-engineer ChatGPT’s AI search.

I wanted a diagnostic engine that would simulate realistic user conversations and capture the complete retrieval layer, so we could monitor the AI prompts and conversations our clients care about. So I built one.

Instead of just knowing whether a brand shows up in a response, I can see exactly what ChatGPT searches for. Our team can use this information to see which sites ChatGPT queries to find information about a client’s vertical, analyze why a competitor was cited instead of a client, and even run tactical tests on client content.

Let’s dive into what I built and my initial findings.

I Built A Script to Watch ChatGPT Search in Real Time

Suganathan’s blog looked at a single query, but I wanted the backend data. Basically, the AI has its own language it speaks when fetching real-time data, and you can see it if you know where to look.

While ChatGPT builds an answer, the website streams data to your browser, including:

What searches it runs
The results that come back
Which of those become citations
Product records

I'll call that information the stream.

Working with AI coding tools, I built a script that types prompts into ChatGPT's website the way a person would. It saves the answer and a screenshot, and it keeps the full stream for every turn.

The other challenge I had to address was Suganthan’s post looked at a single query, but real buyer journeys are a sequence. Most people prompt and then have follow-up questions based on the AI’s answer.

So my script runs an opening question and its follow-ups in the same chat, saving the data at each step. While a single prompt is still a good way to check whether a brand shows up, the follow-ups show what ChatGPT goes looking for once someone starts deciding.

Now my script was ready for testing.

My Test: Three Conversations, Three Turns Each

On October 1, I ran three conversations that I wrote for this test (I chose random topics). Each one happened in a single chat, so every follow-up built on the answer before it. The follow-ups are my guesses at what someone would ask next.

All nine answers came from one run on the free plan, and the stream showed ChatGPT's smaller "mini" model on every turn.

ChatGPT’s Initial Queries Were More Specific Than My Prompts and Even Listed Brands

ChatGPT rewrites your question into its own search queries, and the stream shows them. Here's the first query in each conversation, alongside my original ask and what I was planning to ask for turn number two.

In the first turn, ChatGPT got more specific in its queries than any of the prompts I used, even including specific brands. It also started querying topics and features that I was planning to ask in the second turn.

The luggage query surprised me most. I never mentioned wheels or handles in the first question, yet "durable wheels handles" was in the very first search, one turn before I planned to ask about those features. The cloud backup query did something similar, adding "pricing" a turn before I asked about cost.

I wrote those follow-ups because they seemed like the obvious next questions, and ChatGPT apparently made a similar guess. I can’t tell where that comes from (training data, my wording, or something else) based on the query fan-outs. But the query does give me a written list of what ChatGPT thinks this buyer cares about.

How Did ChatGPT Get to Its Answer?

There was one question I kept coming back to: how did ChatGPT decide which product to recommend?

I used the suitcase conversation as the example here. I wanted to understand why ChatGPT picked the Travelpro Maxlite 5 (its final recommendation), and how it decided that was the most durable option.

Turn 1: ChatGPT Recommends Its Preferred Product

ChatGPT first brought up the Travelpro Maxlite 5 in its response to turn one:

And the initial reasons it gave (turn one) cited Pack & Porter and mentioned price, weight, and value:

The stream doesn't show ChatGPT's reasoning, but it does show what was on the table. ChatGPT ranked the Maxlite 5 first and backed it with two Pack & Porter pages. I read the Pack & Porter's "Best Carry-On Luggage Under $200" page, which ranked the Maxlite 5 first and carried the specs from ChatGPT’s answer:

5.4 lb
A lifetime warranty
Published dimensions of 23 × 14.5 × 9 inches, which is where ChatGPT's warning about strict 22 × 14 × 9 airline limits came from
For the Samsonite Freeform: 6.5 lb, a polypropylene shell, and a 10-year warranty

So the recommendation lines up with the top-cited site. That doesn’t mean Pack & Porter caused the pick: the brand list was already in the query before any page came back.

Turns 2 and 3: ChatGPT Looks to Validate Its Choice

Turn two is where things got interesting. When I asked which suitcase it would choose for durable wheels and handles, ChatGPT ran exactly one search: Travelpro Maxlite 5 wheels handle warranty durable review.

Instead of looking up which carry-on is most durable, ChatGPT looked for support for the product it had already picked. Whatever drove the choice happened before that search.

The sources cited in turn two both have reviews that describe hands-on testing. GearLab ran drop tests (2022). Pack Hacker logged a two-week trip and wrote that the handle "does wiggle substantially" when extended, which shows up in the answer as "can feel somewhat wobbly." The supporting "+1" source, NewCarryOn's Platinum Elite review, relied on OutdoorGearLab's tests rather than its own.

There were two 2026 reviews from the same source set that didn’t get cited, and both included on-page disclaimers that they never tested the bag. A third 2026 review that claimed a year of testing wasn't cited either, so I'm treating this as one data point.

By the third turn on warranty and repair, the search covered a single site. All 13 results came from travelpro.com, and the summary line read "Searched 1 website."

It’s also worth noting that in the data, the first turn's sources already included a Pack & Porter page that ranked carry-ons by warranty.

A Carry-On Question Returned a Checked-Bag Record

One odd thing I noted from the carry-on convo: a product recommended by ChatGPT showed pricing/product disparities. The American Tourister product card showed $174.99, while the answer’s table said about $72. The stream included offers at both prices, but the $174.99 brand offer linked to a large checked bag - not a carry-on. For ecommerce brands, I’d check that the product records match the variant being recommended.

How Early in the Conversation Did ChatGPT Start Citing a Brand Directly?

Turn one of the luggage conversation cited a review site. By the third turn, every citation was Travelpro's own site:

Only the luggage conversation shifted from third-party reviews to the brand. The cloud backup and Spanish conversations went to brand sites from the first turn (and featured a lot of help-center and policy pages):

All 36 citations across those two chats were on the brands' own sites
Half of all the queries in the run (12 of 24) included the word "official"
Two queries used site: to search a single domain

The opening Spanish question did retrieve ten third-party "best Spanish apps" roundups, but all ten appeared only in the second set of results.

If I think about it as a funnel, it makes sense that ChatGPT often turned directly to the brands to answer product-spec questions (such as warranty, cost, and cancellation policies).

Even if your brand isn't cited early in a conversation, you should have help articles, warranty pages, and pricing pages ready for the questions people ask right before they buy.

Every Cited URL Appeared in the First Result Set

The stream recorded an initial set of search results and additional results later. Across all eight turns that searched, every cited URL appeared in the first set. None of the results that appeared only later were cited. I don’t know why, but it gives us another thing to investigate: did our page appear in the set containing the cited sources, or only among the additional results?

Here's the first set for the opening luggage question, in the order the stream numbered it:

Interestingly, five of the nine citations came from one travel gear review site: packandporter.com. This implies that ChatGPT places a high degree of trust in content from Pack & Porter, perhaps because the site explicitly states who writes the reviews, how they rank products, and even how they earn money.

The page titles caught my eye too. The two cited pages are the closest match to what ChatGPT searched for:

Carry-on
Under $200
Travelpro vs Samsonite

A checked-luggage list and a general "luggage under $200" list came back in the same batch and weren't cited. Considering this is a small data set, I'm treating title match as something to test further.

If you track AI visibility, this is worth checking for your own pages. A page can be retrieved in either set, and in this run only one set produced citations.

Where ChatGPT's Answers Diverged from Sources

It Rated Duolingo Without a Corresponding Citation

Duolingo was in the first Spanish query, but ChatGPT never ran a Duolingo-specific search, and no duolingo.com page came back.

The answer still gave Duolingo two stars for speaking practice, and it’s unclear where that rating came from. It could be training data, the conversation so far, or a third-party page that mentions Duolingo. Duolingo does document a speaking feature on its blog, Video Call, which is part of its Max plan, but that page didn't appear in the results.

Surprisingly, ChatGPT's suggested follow-up dropped Duolingo.

On the next turn (turn three), no new search was recorded at all. It picked Pimsleur and started answering in under two seconds.

ChatGPT Ignored Prices That Loaded with JavaScript

The most interesting finding that I want to run further tests on involved how pricing information was listed on a brand’s site.

In the Spanish chat's pricing turn (turn three), the brand with a price in their page's HTML was quoted exactly, while two others had "pricing varies" language.

I checked the source of each pricing page to see what each brand’s pages said and how they were formatted. I did this on October 2, so this information is not necessarily what ChatGPT received:

Pimsleur: The page's HTML includes structured subscription data, "Monthly 7-day free trial" at 19.95, even though the visible text rounds to "$20." The answer quoted "$19.95/month" and the 7-day trial.
Babbel: The prices page loads with almost nothing in it until JavaScript runs ("JavaScript required to see this page"). The answer still quoted Babbel's annual plan ($8.95/month equivalent, $107.40 billed upfront), but said its monthly pricing "varies by offer." Monthly pricing was what I'd asked about.
Busuu: The plans page had no prices in its HTML. The answer said pricing "varies by plan/platform."

There's outside support for why this could happen. Vercel's 2024 study of AI crawlersfound that OpenAI's crawlers download JavaScript files but don't run them, so content that only appears after JavaScript runs may never reach them.

Babbel's annual price still got through from somewhere, so JavaScript content isn't simply invisible. And ChatGPT had already picked Pimsleur a turn earlier without searching, so I can't credit the pricing page for the pick. We've been testing server-rendered HTML against JavaScript-rendered content for clients, and this gives us a clean side-by-side to build on.

CrashPlan's Own Pages Disagreed

Cloud backup had a pricing problem of a different kind. When I asked about cost, ChatGPT searched site:crashplan.com personal backup discontinued individual 2026, checking whether CrashPlan still sells a personal plan.

CrashPlan's help center says its Essential plan is a legacy plan that's no longer available for new purchase. Meanwhile, a CrashPlan Essential page is still live, offering $2.99 a month and a "Start Free Trial" button. ChatGPT quoted my theoretical user, a solo photographer, with the business rate of $7.33 per user per month, and told me to ignore "the older $2.99/month Essential price you may see elsewhere."

For me, this raises a straightforward question: are your pricing pages and help center describing the same offer? If a plan is no longer available to new customers, its old landing page shouldn’t still make it look available. I’d check those pages together, clear up any conflicting information, and rerun the question to see whether the answer changes.

What I'd Test Based on These Findings

None of these have been run yet. Each one comes straight from something in the data above.

You don’t need a script to start digging. Open ChatGPT in Chrome, open DevTools on the Network tab, and ask your category’s main question. Find the response for that conversation and search it for the query ChatGPT wrote. Then ask your two most likely follow-ups in the same chat and note how the queries and cited sites change.

What did ChatGPT add to the first query? Note what it added on its own, such as “durable wheels handles.” Check whether your page answers that concern in plain text. If it doesn’t, add it. Then rerun the same conversation over several weeks and track which set of results your page lands in, as well as whether it gets cited.
Do your help pages answer the follow-up questions? Write down the likely next question in your category, like a warranty claim, a cancellation, or recovering a deleted file. Check that a help article answers it directly, then compare what ChatGPT says against the actual policy.
Does putting the price in HTML change the answer? Take a page where pricing only appears after JavaScript runs. Add plain-text pricing and cancellation terms to the HTML, and leave a similar page unchanged. Then measure whether ChatGPT quotes the exact price or says pricing “varies.”
Which brands does ChatGPT search for? Ask the opening question in fresh chats several times. Record which brands appear in the first query, which get a follow-up query of their own, and which get described without any of their own pages in the results.
Do the product records match the recommendation? For shopping categories, compare the product records ChatGPT carries against the product the answer is recommending. A carry-on question carrying a checked-bag record is the kind of mismatch to look for.
Do cited page titles match the query? For the first set in your category, compare each page’s title to ChatGPT’s query. Over repeated runs, check whether the cited pages are consistently the closest title match. If they are, test retitling a page to match the query wording and see whether it starts getting cited.

Answers vary from run to run, so these tests need repeated captures before and after a change before I’d call something an improvement. Save what you capture so you can check it again later.

Things I Still Don't Know

Whether this repeats. This was one run per conversation, on the mini model, in my own account. The sidebar showed earlier chats on these topics from a first attempt, so I can't rule out account history.
What the model had while writing. The stream shows when results reached the browser, not what was available to ChatGPT behind the scenes.
Which search engine supplies the results. OpenAI says ChatGPT search "sometimes partners with other search providers." Peec's research maps several providers, including Google results gathered by third parties and Microsoft's grounding service. I ran four of the captured queries in Google and Bing myself a day later, from my own browser, so treat this as a rough check: only 3 of 32 Google results and 2 of 19 Bing results matched ChatGPT's first set. Ranking on Google wasn't a good stand-in for being in that set here.
Why a page lands in the first set or the second, and whether being in the first set means the page was read in full.
Whether Pack & Porter's ranking drove the recommendation, or whether both reflect what ChatGPT already believed about these bags.

Where the brand shortlist comes from.

I built this tool to better support the questions our clients ask. Knowing whether a brand shows up in the first answer is useful. Seeing what ChatGPT goes looking for when someone pushes back on that answer gives us much more to work with.

If you want to talk through what this could mean for your AI search strategy, chat with the Seer team.

Closing note: This is a tracker I'm building as I go, and what ChatGPT exposes keeps shifting. I keep records of the raw streams so I can go back and recheck patterns like the citation split against older captures. I can also monitor the parts we don't understand yet, and rerun fixed prompts to see what changes.

Methodology

I captured three conversations on October 1, 2026, each with three prewritten prompts asked in the same chat, for nine prompts in total. They ran on the ChatGPT free plan in a desktop Chrome browser, and every turn was answered by the mini model. For each turn, we saved the answer, a screenshot and the full stream. Counts come from the stream itself. Pages were fetched on October 2, the day after the capture, and may differ from what was live during the run. None of the proposed tests have been run.

Agentic Commerce May Break Your Measurement Before It Breaks Your Strategy.

Admin — Wed, 28 Oct 2026 19:01:08 GMT

TL;DR

AI agents can complete a purchase without ever loading your site, so GA4 may never record the sale while your order system does.

Same Prompt, Different Data Structure: ChatGPT's Search Data Keeps Changing

Admin — Thu, 22 Oct 2026 19:00:00 GMT

TL;DR

I ran the same ChatGPT prompt multiple times and saw the search data change across weeks, days, and even minutes
Between Sept. 3 and Sept. 16, ChatGPT's underlying pipe-language data dropped from 4 lines to 0, while unique retrieved URLs went up (from 14 to 40)
Two identical runs 8 minutes apart on Sept. 17 shared zero exact query strings - the same prompt produced different search behavior
This tells us that measuring AI search visibility through single-run prompts is misleading and ineffective
It also shows the value of looking at the full retrieval process (not just the rendered answer) when reporting on AI brand visibility: 34 of the 40 URLs retrieved on Sept 16 were retrieved but not cited

Skip to the methodology →

Competitive Analysis for Paid Media: How To Find the Broken Stage

Admin — Wed, 07 Oct 2026 19:01:13 GMT

TL;DR

Competitive research can show which stage of your funnel is broken (Be Seen, Be Believed, or Be Chosen) and which channel to fix it in.

Re-imagining AI Visibility KPIs in a Query Fan Out world

Admin — Fri, 02 Oct 2026 16:30:48 GMT

Wil, take me right to the KPI for long term sustainable GEO visibility!!

SaaS Marketers - Promote your Claude & Codex connectors & win in AI search

Admin — Mon, 28 Sep 2026 20:30:29 GMT

Claude and I had a 12,000 word back and forth on how to use Cloudflare and GA4 data to answer a hard question. If you were a SaaS marketer what would you do with all that context to make better answers?

AI Citations Aren't Your Top KPI: ChatGPT-5.6 Luna Runs More Targeted Searches

Admin — Thu, 24 Sep 2026 19:36:11 GMT

TL;DR

We ran a direct comparison of 346 matched prompts to see whether ChatGPT-5.6 Luna behaved differently from GPT-5.5
GPT-5.6 Luna used web search more often but ran fewer queries, resulting in little movement on the number of fan-out queries compared to GPT-5.5
Domain-target searching increased drastically (more than 3x), while consulted sources and inline citations dropped
Our take: Luna is getting better at finding information and needs fewer resources, so brands should pay more attention to the full retrieval process to measure AI search visibility

How Do ChatGPT-5.5 and GPT-5.6 Differ?

We’ve researched ChatGPT-5.5 extensively, including a deep dive into fan-out patterns. Now that GPT-5.6 Luna is the default model as of July 9, 2026, it's the model people get when interacting with ChatGPT and the model brands will be reporting on through AI visibility trackers.

With that in mind, we wanted to know how the two models compared. Specifically:

How did the retrieval process change across domains and citations?
Does GPT-5.6 Luna search similarly, or were there notable differences?
Did fan-out query behavior shift or stay the same?

We compared GPT-5.5 and GPT-5.6 Luna across 346 matched prompts from six industries.

Spoiler alert: At first glance, the two models looked very similar. Across the 346 matched prompts, they ran almost exactly the same number of fan-out queries. The difference is how targeted Luna's searches are. It used the site: operator in 76% of its fan-out queries, up from 22% with GPT-5.5.

That leaves far fewer open searches for everyone else. So if Luna isn't already searching your site directly, your barrier to showing up is higher than it used to be. Check out our top findings below.

1. GPT-5.6 Luna Uses Web Search More Often, But Runs Fewer Queries Per Search

GPT-5.6 Luna turned to web search more often than GPT-5.5. Across the six industries, Luna searched on 85.2% of calls, up from 80.8%. But each time it searched, it ran fewer fan-out queries: 3.42 per search, down from 3.63.

Those two changes almost exactly cancel each other out. Luna searched more often but ran fewer queries each time, so the total number of fan-out queries barely moved. GPT-5.5 ran 733 and Luna ran 729, a difference of just four.

We thought there would be more fan-out queries after the model update, especially given the increase we saw when GPT-5.5 was released. Instead, the total stayed the same, and the change showed up in how Luna searched.

Our hypothesis is that Luna needs fewer queries per search because its queries are more targeted. When a fan-out query goes straight to a specific website (more on that in the next finding), one query can do the job that used to take several broader ones. So the two changes balance out in volume, but not in effect. More prompts now send ChatGPT to the web, and each search covers less ground.

2. Domain-Targeted Searching More Than Tripled, Especially on Non-Brand Prompts

GPT-5.5 had already begun using the site: operator (a search command that limits results to a single website) in fan-out queries to target specific brand sites for answers and, in our opinion, prevent low-quality responses. GPT-5.6 Luna elevated this retrieval method even further.

GPT-5.5 conducted site: searches in about 22% of its fan-out queries, while Luna did so in 76%.

On branded prompts, third-party sites (publishers, review sites, research orgs, etc.) appeared at roughly the same rate with both models: 26.6% with GPT-5.5 and 24.4% with Luna. So even when a user asks about your brand directly, ChatGPT still looks beyond your website.

That means the perception of your brand outside of your website matters just as much as the narrative you share on sources you control.

The actual jump came from prompts that didn't mention a brand. Non-brand prompts saw a rise in site: searches, from 20.6% in GPT-5.5 to 77.3% in GPT-5.6. That means Luna was the one to decide on a domain to search even when the user didn’t specify a brand or website.

Since ChatGPT is getting more targeted with its retrieval, the overall number of consulted sources (the sources ChatGPT pulls in while searching, before it decides which ones to cite) fell 43%, from 6,134 with GPT-5.5 to 3,481 with Luna.

This presents a unique opportunity for brands launching AI search programs. If GPT-5.6 Luna is already targeting your site with site: searches, make sure the information it's looking for is easy for both people and AI to find.

If your brand isn't one of the sites Luna searches directly, you may see fewer citations now that it's the default model. Open searches, the ones any site could appear in, fell from 575 to 175, so getting results will likely take more work.

Look at every stage of retrieval, from fan-out queries to consulted sources to citations, to see how and where ChatGPT finds information about your brand.

3. Inline Citations Fell 13%

We know Luna searched in a more targeted way and consulted fewer sources. In the final answers, it also gave fewer inline citations (the sources linked in the answer itself).

Across all 346 matched prompts, GPT-5.5 produced 2,054 inline citations, while Luna produced 1,790, a decrease of 12.9%. Searching more often didn't translate into more citations.

Our take is that Luna has gotten better at knowing where to find information, so it needs fewer sources to back up its answers.

This matters for any AI search program that uses citations as a KPI or as the success metric for an experiment. If your citations dropped after Luna became the default, that may have nothing to do with your strategy or the tests you have running.

Think of it like a Google algorithm update: the platform changed, and performance shifted with it. Mark the switch to Luna in your reporting so a platform-wide drop doesn't get read as a failed experiment.

So What Should You Do Now?

How can you apply these findings to your AI search strategy? Here are my recommendations:

Look further down the retrieval process when you’re measuring AI search visibility. Most AI visibility trackers only report citations, the bottom line of the retrieval process.

The bigger benefit comes from the layers behind them: the fan-out queries ChatGPT runs, the domains it targets with site: searches, and the sources it consults before deciding what to cite. That's where you'll see why your citations moved, not just that they did.

Pay closer attention to the connections the models appear to make between brands and topics. Luna often decides which sites to search before it retrieves anything, so being recognized as a relevant source for a topic matters before any page gets retrieved.

Run a base set of nonbrand prompts you’ve been monitoring to capture the fanout queries and sort the site: searches by topic. Which topics does ChatGPT search your domain for, and which does it search your competitors' domains for instead?

Compare across models or over time, and you'll see where you're gaining ground and where a competitor owns the association. The topics your competitors own are where you should focus next: both in your own content, and in your coverage on publishers, review sites, and research organizations.

Ready to level up your AI search strategy but don’t know where to start? The Seer team has extensively researched what works and what’s a short-term trend, and we can recommend where to put your resources based on your industry, budget, and team size. Chat with us to learn more.

Our Most Valuable Blogs of 2026: According to Our Readers

Hannah Cooley — Tue, 22 Sep 2026 21:00:18 GMT

What's my favorite horror movie? When I patted myself on the back for setting up an ~autonomous workflow~ that kept our newsletter subscriber hygiene clean by automatically unsubscribing anyone who hasn't opened any of our emails in the past 12 months.

Google Is Not Going to Stop Ranking Your Claude-Assisted Content

Admin — Fri, 04 Sep 2026 18:24:31 GMT

TL;DR

Claude’s watermark only flags that Claude was “likely involved,” so it can’t tell a fully AI-drafted piece from a human draft Claude lightly edited
It’s an EU compliance requirement (Code of Practice on Transparency of AI-Generated Content). This is not a new Google ranking signal.
Google has run its own version (SynthID-text)on Gemini for two years and never used it to rank anything
The real traffic risk is thin, templated content and not the watermark. Sites that scaled content off a copy-the-template approach saw more than half those lose 30%+ of peak organic traffic
What still matters: unique inputs and a human accountable for the final piece

Anthropic started watermarking Claude's text output.

AI Visibility Is Lying to You: There’s No One-Size-Fits-All AI Search KPI

Admin — Mon, 31 Aug 2026 19:59:48 GMT

TL;DR

Everyone measuring AI search is defaulting to the same KPIs, but there is no one-size-fits-all AI search KPI
New research shows 54% of AI search categories lack a clear category owner, and those categories represent 89% of all AI search demand
If someone else owns your category, you should stop chasing visibility and instead focus on accuracy: when you do show up, does the model describe you the way you actually want to be described?
Accurately measuring AI search KPIs requires a brand canon, a living document of your brand’s claims, positioning, facts, and more

Your Visibility Score Can Be Perfect and Still Be Wrong

AI can mention a brand in every prompt you track, but that brand can still be losing.

How we built an agentic loop to improve every skill in our company, instantly

Admin — Thu, 20 Aug 2026 18:48:49 GMT

TL;DR

I reviewed our traffic-decline-analysis skill's and found four real problems: a traffic number from the wrong source, a real drop mislabeled "stable," no way to trace where a number came from, and no long-term trend view
I gave that feedback in a 9-minute recording and an agent picked up my feedback and fixed it the same day, and it shipped to everyone at Seer.
I found another problem the skill was picking which checks to run instead of running all of them, every time
An offhand idea from that same review became a new feature days later. The tool now runs off a simple spreadsheet export instead of needing live client data

Here’s how it all happened…

What Is AI Saying About Your Brand? Develop a Brand Canon and Find Out.

Admin — Mon, 17 Aug 2026 16:37:16 GMT

TL;DR

Brands are measuring AI visibility without tracking if the perception of their brand is even accurate
We recommend building a brand canon, which tracks facts, differentiators, brand claims, what your brand wants to be known for, and what you don’t
Once you’ve created your canon, you can use it to compare what you want AI to say about your brand versus what it actually says (and fix the output by tracing claims to their source)
Learn how to build your own brand canon, how to measure the accuracy of claims made by AI, and how to prioritize fixing falsehoods and inaccuracies

What Happens if AI Shares Inaccurate Info About Your Brand?

A potential customer asks AI what your company does, and the answer comes back about 70% right (per Seer's analysis of 28,123 responses about our brand across 6 AI models). But nearly 30% of the information is wrong or outdated, or the model declines to answer.

Our Early View of ChatGPT Ads: 5 Initial Observations - Updated July 28, 2026

Admin — Tue, 28 Jul 2026 04:00:00 GMT

When we first published this on July 15, we were working from three days of ChatGPT ads data. Revisiting the data this week, our main takeaways haven't changed, but ecosystem is evolving. Share of responses with ads rose significantly, a wider set of brands is now buying them, and early patterns are sharpening.

The updated numbers below now reflect about 2 weeks of data. With that, we advise you treat this as a developing study. It's still early and we are prepared for this to change fast.

OpenAI officially launched the ChatGPT Ads pilot on February 9, 2026, rolling the placements out more widely as of May. While we've been tracking how ads work, tests and usability through our own pilot and other studies, we have not had a view into how often they show up, what types of prompts they’re shown for, and how well they align with those responses that they are served under.

That changed this week when ads started appearing in our AI tracker.

How to Use AI to Enhance Client Relationships (Without Losing Your Competitive Edge)

Admin — Fri, 10 Jul 2026 04:00:00 GMT

This blog was originally published on December 4, 2023 and was updated on July 10, 2026.

Is Your Brand Considering a Reddit Strategy for AI Visibility? Here’s What to Do.

Admin — Wed, 08 Jul 2026 17:11:36 GMT

Brands have heard that Reddit matters for AI visibility. Many of them are adopting the mindset of “we need to be on Reddit” without thinking the strategy through.

Can a prompt undo years brand building? Our AI interview tests say yes

Admin — Wed, 01 Jul 2026 17:11:04 GMT

TL;DR: We interviewed 28 people across almost 100 tasks, having them use AI to solve a product selection problem, and we ask them what brands people intend to buy, before using AI, then we check what brand they considered after.

One AI research session cut GE's purchase consideration in half while lesser known brands like Rheem surged.

The questions we wanted to answer:

When someone has a task, what brands do they think would solve their problem before they search or use AI?

The Proof: ChatGPT 5.5's fanout patterns reveal the importance of brand

Admin — Thu, 25 Jun 2026 17:04:44 GMT

If you've been looking at the whole "do I have to do listicles" thing and throwing up in your mouth a little, this post is for you.

Your AI Traffic Is Hiding: A Practical Guide to Tracking ChatGPT, Claude, and More

Admin — Mon, 22 Jun 2026 04:00:00 GMT

This blog was originally published on December 2, 2024 and was updated on June 22, 2026.

⚠️ AI Tracking Is A Moving Target

The AI tracking landscape evolves rapidly. Since our original publication, Google launched its native AI Assistant channel (May 2025), ChatGPT shifted from 'citations' to 'sources,' Claude added embedded links capabilities, and ChatGPT introduced advertising. Methods described in this blog represent current best practices as of the last update date, but we encourage you to validate these approaches yourself as platform behaviors continue to change.

Are you trying to better understand how searchers are using ChatGPT, Gemini and other AI platforms to find your websites? We sure are!