Introduction: Taking Stock at the End of 2025

Over the last few years, I’ve been very active on LinkedIn, sharing observations from the front lines of enterprise SEO. I’ve spent a large part of my career building and managing SEO programs on the enterprise side where scale, complexity, and cross-functional execution are unavoidable realities.

More recently, I moved to seoClarity, where I’m responsible for building our AI Search Visibility Monitoring and Optimization Platform, ArcAI. That shift from running SEO programs to building tooling for the next generation of search has given me a unique vantage point.

As 2025 comes to a close and we look ahead to 2026 and beyond, I’ve been trying to take stock of a few things:

• Where is AI search actually today?

• How is it changing how customers discover, evaluate, and choose brands?

• What do businesses fundamentally misunderstand about this shift?

• And why, despite all the noise, is SEO still the right team to own AI search?

This writeup is an attempt to answer those questions not tactically, but strategically.

The Quiet Collapse of the Top of the Funnel

Let’s start with the uncomfortable truth.

Top-of-funnel traffic is disappearing.

Not because demand is gone but because it’s being resolved before users ever reach a website.

AI search platforms like ChatGPT, Perplexity, Claude, and Google’s AI experiences are increasingly:

• Answering questions directly

• Comparing options

• Explaining trade-offs

• Recommending solutions

In many cases, the user completes most of their decision-making inside the AI interface itself.

What does that mean for businesses?

It means:

• Fewer visits

• Fewer measurable touchpoints

• Less visible attribution to Organic Search

• And yet higher intent when users finally do arrive

I’ve seen this pattern repeatedly: AI-driven traffic converts better, because users only click when they’re ready to act.

AI didn’t kill the funnel.
It compressed it.

The New Content Paradox: Authority Without Traffic

This leads to the first major dilemma businesses will face in 2026:

You still need to produce content even when that content may never generate traffic.

Why?

Because AI search engines don’t rank pages they retrieve knowledge.

If your content:

• Builds topical authority

• Clearly explains concepts

• Defines products, features, and use cases

• Answers nuanced questions

…then you have a higher chance of being retrieved, mentioned, cited, and trusted when AI generates an answer.

The business challenge is obvious:

• How do you justify content investment without traffic?

• How do you measure ROI?

• How do you scale content production to the volume AI retrieval requires?

The answer is not “more writers.”

The answer is automation, systems, and a fundamentally new skill set.

Why Traditional Content Workflows Break

The old model looks like this:

Content Brief → Content Outline → Writer → Optimize → Publish

That model does not scale to:

• Hundreds of thousands of questions

• Multiple personas

• Dozens of decision stages

• Continuous AI-driven retrieval

• Continuous Content Refresh

Modern content production needs:

• A source of truth for the business

• Structured knowledge about products, personas, pain points, and use-cases

• Human-in-the-loop automation where 80–90% is system-generated and reviewed by experts

Content is no longer a campaign output.
It’s infrastructure.

AI Search Still Runs on SEO Foundations

There’s a misconception that AI search makes SEO irrelevant.

In reality, the opposite is true.

While large language models rely on internal knowledge for simple queries, complex queries trigger retrieval workflows often involving:

• Query expansion (fan-outs)

• External content retrieval

• Source evaluation and citation

This is where traditional SEO foundations still matter:

• Crawlability

• Indexability

• Content structure

• Internal linking

• Authority signals

SEO is the backbone of AI retrieval.

The difference is that the output is no longer a ranking it’s an answer.

The Real Challenge: Query Fan-Outs You Can’t See

One of the hardest problems in optimizing for AI search is this:

You don’t (reliably) know what queries the AI is expanding into.

Most of these queries:

• Don’t exist in keyword tools

• Will never show up in Search Console(except from Google AI mode, in certain cases)

• Are dynamically generated by AI systems

This is why prompt research becomes essential.

Not prompt guessing but structured modelling of:

• Customer intent

• Customer personas

• Business use cases

• Decision paths

You cannot track millions of prompts.
But you can build a representative sample that reflects how your audience thinks.

That sampling problem is one of the most important new skills in modern SEO(or AEO/GEO if you want to call it that).

Log Files: The Most Underrated AI Search Signal

If I had to point to the most valuable data source for understanding and actioning on AI search today, it wouldn’t be rankings or traffic.

It would be log files.

Log files show you:

• Which AI bots are hitting your site

• What pages they’re accessing

• How frequently

• And how that behavior differs from human traffic

In the future, SEO teams will need to:

• Identify LLM training bots

• Separate retrieval bots from agent bots

• Correlate bot activity with downstream visibility and conversions

The companies that win will be the ones that connect:

AI bot interaction → retrieval → citation → business outcome

Optimization in 2026: From Pages to Questions

Optimization workflows are changing in subtle but profound ways.

One of the most important steps almost no one talks about is this:

Generate synthetic questions and validate whether your website actually answers them.

Think about prompt research in layers:

1. The universe of questions your customers could ask

2. A large synthetic set to test content coverage

3. A small, high-impact subset you actively track for AI Search visibility

If your site doesn’t answer a question:

• You lose retrieval possibility

• You lose narrative control

• And AI fills the gap from somewhere else

Sometimes that “somewhere else” is a third-party site which may be acceptable.
Other times, it’s inaccurate or misleading which is dangerous.

Accuracy and Sentiment Are the New Brand Risks

Being mentioned by AI is not enough.

Businesses now have to ask:

• How are we described?

• How are we mentioned?

• Are we cited correctly?

• Is the sentiment accurate?

• Which attributes are emphasized or ignored?

If AI search gets your pricing, positioning, or capabilities wrong, users may never give you a chance to correct it.

This creates an entirely new operational need:

• Detect inaccuracies

• Identify the source (your site vs third parties)

• Correct them through content and influence

SEO becomes narrative governance.

Why SEO Is the Right Team to Own AI Search

This brings me to what I believe is the most important organizational insight.

AI search doesn’t belong to:

• Paid media

• Brand

• PR

• Or product alone

It belongs to SEO expanded and evolved.

Modern SEO teams already sit at the intersection of:

• Content

• Technology

• Analytics

• User intent

• Cross-functional execution

But the charter needs to change.

SEO’s New Responsibilities

SEO teams must evolve to own:

• Search visibility (where and when the brand appears)

• Citations, Share of Voice

• Accuracy and misinformation risk

• Sentiment and narrative by intent

• Retrieval readiness and accessibility

This often requires new capability lanes:

In practice, this often looks like an SEO-led visibility control tower working across marketing, product, engineering, analytics, and communications.

SEO doesn’t do everything.
But it coordinates everything.

Rethinking Metrics and Attribution

Traditional organic KPIs are no longer sufficient.

Traffic alone cannot explain:

• Influence

• Visibility

• Or lost demand

The future measurement stack includes:

• AI Search visibility

• Brand mentions and citations

• Accuracy scores

• Sentiment by attribute

• Conversion performance of AI-referred traffic

• Log-based interaction data

Attribution has always been hard for SEO.
Now it’s unavoidable.

Looking Ahead: Preparing for AI-Native Experiences

AI platforms are rapidly adding:

• Shopping research

• Deep research

• Task automation

• Agent-based workflows

As these systems learn what users are trying to accomplish, they will build vertical-specific experiences.

To participate in those experiences, brands need:

• Deep contextual content

• Clear knowledge representation(think OpenAI’s product feeds as a starter)

• Accessible and accurate information

AI doesn’t integrate “pages.”
It integrates understanding.

A Final Thought for Business Leaders

If there’s one idea I want to leave you with, it’s this:

In the AI era, the brands that win won’t be the ones with the most content but the ones whose businesses are best understood by machines.

SEO is no longer about traffic.

It’s about:

• Visibility without clicks

• Authority without rankings

• Trust without touchpoints

And that makes SEO more strategic not less than it has ever been.

Query Context: User asked “What are the best sunscreens for Oily Screen to buy in Toronto?”
Response Type: Product-focused search with e-commerce integration
Total Lines: 296
Key Features: Product entities, real-time shopping data, location-specific recommendations, content moderation

METHODOLOGY EXPLANATION

Throughout this document:

• RAW DATA = Exact content from the SSE stream response
• TECHNICAL ANALYSIS = My interpretation and explanation of what’s happening
• SEO INSIGHTS = Implications for search and AI understanding

PHASE 1: PROTOCOL INITIALIZATION (Lines 1-3)

Line 1-3: Delta Encoding Setup

RAW DATA:

1→event: delta_encoding
2→data: "v1"
3→

TECHNICAL ANALYSIS:

• Line 1: event: delta_encoding – Establishes the SSE event type as delta encoding system
• Line 2: data: "v1" – Specifies version 1 of the delta encoding protocol
• Line 3: Empty line – Standard SSE protocol delimiter between events

SEO INSIGHTS: Delta encoding allows incremental updates, reducing bandwidth and enabling real-time search result streaming

PHASE 2: MESSAGE CHAIN INITIALIZATION (Lines 4-9)

Lines 4-6: System Message Creation

RAW DATA:

4→event: delta
5→data: {"p": "", "o": "add", "v": {"message": {"id": "81cb190f-1f53-4f49-a732-6f969b300329", "author": {"role": "system", "name": null, "metadata": {}}, "create_time": null, "update_time": null, "content": {"content_type": "text", "parts": [""]}, "status": "finished_successfully", "end_turn": true, "weight": 0.0, "metadata": {"is_visually_hidden_from_conversation": true, "model_switcher_deny": []}, "recipient": "all", "channel": null}, "conversation_id": "68909c60-03e8-8003-812d-57045fb81d7d", "error": null}, "c": 0}     
6→

TECHNICAL ANALYSIS:

• Delta operation: "p": "" (path), "o": "add" (operation), "v": {...} (value)
• Message ID: 81cb190f-1f53-4f49-a732-6f969b300329 – Unique identifier for system message
• Author role: "system" – Internal system message, not user-visible
• Status: "finished_successfully" – System setup completed
• Weight: 0.0 – System messages have no conversation weight
• Metadata: "is_visually_hidden_from_conversation": true – Hidden from user interface
• Conversation ID: 68909c60-03e8-8003-812d-57045fb81d7d – Session identifier
• Counter: "c": 0 – First message in sequence

SEO INSIGHTS: System establishes conversation context before processing search queries

Lines 7-9: User Message Registration

RAW DATA:

7→event: delta
8→data: {"v": {"message": {"id": "738a36e6-7102-436b-989e-96854935eca3", "author": {"role": "user", "name": null, "metadata": {}}, "create_time": 1754307685.038, "update_time": null, "content": {"content_type": "text", "parts": ["What are the best sunscreens for Oily Screen to buy in Toronto?"]}, "status": "finished_successfully", "end_turn": null, "weight": 1.0, "metadata": {"selected_github_repos": [], "serialization_metadata": {"custom_symbol_offsets": []}, "request_id": "969dc9188d2fac9f-DFW", "message_source": null, "timestamp_": "absolute", "model_switcher_deny": []}, "recipient": "all", "channel": null}, "conversation_id": "68909c60-03e8-8003-812d-57045fb81d7d", "error": null}, "c": 1}   
9→

TECHNICAL ANALYSIS:

• Message ID: 738a36e6-7102-436b-989e-96854935eca3 – Unique user message identifier
• Create time: 1754307685.038 – Unix timestamp (January 14, 2025, 10:28:05 GMT)
• User query: "What are the best sunscreens for Oily Screen to buy in Toronto?" – Note: typo “Oily Screen” vs “Oily Skin”
• Weight: 1.0 – Full weight for user messages
• Request ID: 969dc9188d2fac9f-DFW – Internal request tracking (DFW = Dallas Fort Worth data center)
• Counter: "c": 1 – Second message in sequence

SEO INSIGHTS: User query contains product intent, location specificity, and skin type targeting

PHASE 3: SEARCH INTENT CLASSIFICATION (Lines 10-12)

Lines 10-12: Sonic Classifier Analysis

RAW DATA:

10→event: delta
11→data: {"v": {"message": {"id": "5473d6ed-5f74-4e7a-80d4-69c9ed665eaa", "author": {"role": "assistant", "name": null, "metadata": {}}, "create_time": 1754307685.3786674, "update_time": null, "content": {"content_type": "code", "language": "unknown", "response_format_name": null, "text": "search(\"What are the best sunscreens for Oily Screen to buy in Toronto?\")"}, "status": "finished_successfully", "end_turn": null, "weight": 1.0, "metadata": {"sonic_classification_result": {"latency_ms": 22.100909001892433, "search_prob": 0.682043286148492, "complex_search_prob": 0.0, "force_search_threshold": 0.65, "classifier_config_name": "sonic_classifier_ev3", "search_complexity": "simple"}, "citations": [], "content_references": [], "request_id": "969dc9188d2fac9f-DFW", "message_type": "next", "model_slug": "gpt-4o", "default_model_slug": "auto", "parent_id": "738a36e6-7102-436b-989e-96854935eca3", "model_switcher_deny": []}, "recipient": "web", "channel": null}, "conversation_id": "68909c60-03e8-8003-812d-57045fb81d7d", "error": null}, "c": 2}    
12→

TECHNICAL ANALYSIS:

• Content type: "code" – Search function call representation
• Search function: search("What are the best sunscreens for Oily Screen to buy in Toronto?") – Direct query passthrough
• Sonic classifier results:
  • Latency: 22.1ms – Classification processing time
  • Search probability: 68.2% – Confidence this requires web search
  • Complex search probability: 0.0% – Classified as simple search
  • Threshold: 65% – Minimum probability to trigger search
  • Classifier version: "sonic_classifier_ev3" – Third evolution of Sonic classifier
  • Search complexity: "simple" – Non-complex search pattern
• Recipient: "web" – Routed to web search system
• Model: "gpt-4o" – OpenAI GPT-4 Omni model

SEO INSIGHTS: AI determines search intent with 68.2% confidence, triggering automated web search for product recommendations

PHASE 4: SERVER INFRASTRUCTURE METADATA (Lines 13-14)

Lines 13-14: Performance & Infrastructure Status

RAW DATA:

13→data: {"type": "server_ste_metadata", "metadata": {"conduit_prewarmed": true, "fast_convo": true, "warmup_state": "warm", "is_first_turn": true, "model_slug": "gpt-4o", "did_auto_switch_to_reasoning": false, "auto_switcher_race_winner": null, "is_autoswitcher_enabled": false}, "conversation_id": "68909c60-03e8-8003-812d-57045fb81d7d"}
14→

TECHNICAL ANALYSIS:

• Server metadata type: "server_ste_metadata" – Server state event metadata
• Conduit prewarmed: true – Infrastructure connection already established
• Fast conversation: true – Optimized conversation pipeline active
• Warmup state: "warm" – Server resources ready
• First turn: true – Initial user interaction in conversation
• Auto-switching: All disabled – No automatic model switching performed

SEO INSIGHTS: Infrastructure optimization ensures low-latency search responses for commercial queries

PHASE 5: WEB SEARCH EXECUTION & RESULTS (Lines 15-97)

Lines 15-17: Search Process Initiation

RAW DATA:

15→event: delta
16→data: {"v": {"message": {"id": "3b417387-5632-4fb9-9f22-6aaf3b0d9b6a", "author": {"role": "tool", "name": "web.run", "metadata": {"sonicberry_model_id": "current_sonicberry_unpaid_oai", "real_author": "tool:web", "source": "sonic_tool"}}, "create_time": 1754307686.8143666, "update_time": null, "content": {"content_type": "text", "parts": [""]}, "status": "in_progress", "end_turn": false, "weight": 1.0, "metadata": {"search_result_groups": [{"type": "search_result_group", "domain": "nowtoronto.com", "entries": [{"type": "search_result", "url": "https://nowtoronto.com/lifestyle/health/best-sunscreen-for-face/", "title": "The 13 best face sunscreens - NOW Magazine", "snippet": "", "ref_id": null, "pub_date": null, "attribution": "nowtoronto.com"}]}], "debug_sonic_thread_id": "thread_68909c66d7f880038ede6a66aea7b624", "search_turns_count": 1, "search_source": "composer_auto", "client_reported_search_source": "composer_auto", "request_id": "969dc9188d2fac9f-DFW", "message_type": "next", "model_slug": "gpt-4o", "default_model_slug": "auto", "parent_id": "5473d6ed-5f74-4e7a-80d4-69c9ed665eaa", "model_switcher_deny": []}, "recipient": "all", "channel": null}, "conversation_id": "68909c60-03e8-8003-812d-57045fb81d7d", "error": null}, "c": 3}  
17→

TECHNICAL ANALYSIS:

• Tool name: "web.run" – Web search execution tool
• Sonicberry model: "current_sonicberry_unpaid_oai" – OpenAI’s unpaid search model
• Status: "in_progress" – Search actively running
• First result: NOW Toronto article on face sunscreens
• Search metadata:
  • Thread ID: Debug identifier for search thread
  • Turn count: 1 – First search iteration
  • Source: "composer_auto" – Automatically triggered search

SEO INSIGHTS: System immediately finds local Toronto content, demonstrating location-aware search capabilities

Lines 18-97: Progressive Search Result Accumulation

TECHNICAL ANALYSIS: Lines 18-97 show the system progressively collecting search results using delta updates. Each new result is appended to the growing list:

Key Search Results Identified:

Lines 18-20: fluidmediallc.com – “Best Sunscreens for Oily Skin in 2025”
Lines 21-23: straight.com – “Top 5 Best Mineral Sunscreens in Canada 2025”
Lines 24-26: thesun.co.uk – Doctor’s guide to sunscreens for oily skin (with detailed snippet)
Lines 27-29: marieclaire.co.uk – Best mineral sunscreens with product recommendations
Lines 30-32: skinsort.com – “8 Best Oil Free Sunscreens From Canadian Brands”
Lines 33-35: reddit.com – User discussion on water-based sunscreens
Lines 36-38: dailyhive.com – “We tried the best sunscreens in Canada”
Lines 39-41: nypost.com – Korean sunscreen guide with dermatologist insights
Lines 42-43: reddit.com/r/CanSkincare – “Looking for an good sunscreen for Acne/oily skin” (Canadian skincare community, posted Jan 15, 2025)
Lines 46-47: instyle.com – “I Tried a Dermatologist-Approved Summer Skin Care Routine-Now I Look 10 Years Younger” (comprehensive summer skincare guide with SPF recommendations including Dermalogica BioLumin-C, Dr. Whitney Bowe retinols, and SkinCeuticals serums, posted Jul 30, 2025)
Lines 48-49: reddit.com/r/SkincareAddiction – “Need the best sunscreen for oily skin, recommendations please! [Product Request]” (popular skincare community discussion, posted Aug 22, 2024)
Lines 52-53: rggregory.com – “The 9 Best Sunscreens For Oily Skin, According To Dermatologists (2025)” (dermatologist-curated list for oily skin types)
Lines 54-55: reddit.com/r/MadeInCanada – “Canadian Sunscreens” (discussion about Canadian-made sunscreen alternatives, posted Jan 11, 2025)
Lines 57-58: straight.com – “Top 6 Best Sunscreens in Canada 2025” (second Straight.com article focusing on general sunscreen recommendations)
Lines 60-62: gq.com – “The Best Face Sunscreen Prevents Burns Now and Skin Damage Later” (men’s lifestyle magazine guide with dermatologist Dr. Asmi Sanghvi recommendations including Coola Face SPF 50, Bask Daily Invisible Gel SPF 40, and EltaMD UV Clear, posted Jun 27, 2024)
Lines 63-65: vogue.com – “The Best Sunscreen for Face Recommended for Year-Round Protection” (authoritative beauty guide with 7 top recommendations including SkinCeuticals Physical Fusion SPF 50, EltaMD UV Clear SPF 46, Paula’s Choice Hydrating Fluid SPF 50, posted Aug 8, 2024)
Lines 67-68: nypost.com – “Glow, guard and go: Shop IMAGE Skincare’s all-in-one SPF solution” (IMAGE Skincare DAILY PREVENTION collection featuring SPF products with XOSM Technology, posted Jul 28, 2025)
Lines 69-71: thetimes.co.uk – “The best luxury face SPFs” (luxury beauty guide featuring high-end options like 111SKIN, Augustinus Bader, Dr. Barbara Sturm Sun Drops, Emma Lewisham mineral SPF, Sisley, La Mer, posted Jan 14, 2025)
Lines 72-74: realsimple.com – “Dermatologists Trust These Tinted Sunscreens for Lightweight Coverage and Heavy-Duty SPF” (comprehensive tinted sunscreen guide featuring EltaMD UV Clear SPF 46, CeraVe Hydrating Mineral SPF 30, Eucerin Sun Tinted Mineral, priced $14-$50, posted Jul 30, 2025)
Lines 75-77: prevention.com – “19 Best Mineral Sunscreens for Face and Body, According to Dermatologists and Editor Testing” (extensive mineral sunscreen testing guide featuring La Roche-Posay Anthelios Mineral Ultra-Light Fluid SPF 50, CeraVe Mineral Sunscreen SPF 50, Supergoop! Mineral Unseen Sunscreen, posted Jan 14, 2025)
Lines 78-80: glowandbloom.ca – “Bye-Bye Breakouts! – The Best Sunscreens for Oily, Acne-Prone Skin” (Canadian beauty blog specializing in acne-prone skin solutions)
Lines 81-83: straight.com – “Top 8 Best Oil-Free Sunscreens in 2025” (third Straight.com article specifically targeting oil-free formulations)
Lines 84-86: verywellhealth.com – “Which Facial Sunscreens Do Dermatologists Actually Recommend?” (medical authority site with dermatologist recommendations including Eucerin Sun Face Oil Control Tinted SPF 50+, Glo Skin Beauty Oil-Free SPF 40+, Charlotte Tilbury Invisible UV Flawless Primer SPF 50, posted Jul 6, 2023)
Lines 87-89: cosmopolitan.com – “11 Best Face Sunscreens of 2025, Vetted by Cosmo Editors” (beauty magazine editorial picks for 2025)
Lines 90-92: glamour.com – “12 Best Sunscreens for Oily Skin 2025, Tested and Reviewed” (beauty magazine testing and review of oily skin specific products)
Lines 93-95: reddit.com/r/45PlusSkincare – “Sunblock for us oily babes” (mature skincare community discussion about sunscreen for oily skin, posted Jul 13, 2024)

Pattern Analysis:

• Each result uses "p": "/message/metadata/search_result_groups" path with "o": "append" operation
• Results are grouped by domain for better organization
• Some results include detailed snippets, others are title-only
• Mix of commercial, editorial, and user-generated content

SEO INSIGHTS:

• System prioritizes authoritative health/beauty sources
• Location-specific Canadian content heavily featured
• Product-focused content dominates results
• Reddit discussions included for real user experiences

Lines 96-97: Search Completion

RAW DATA:

96→event: delta
97→data: {"p": "/message/status", "o": "replace", "v": "finished_successfully"}

TECHNICAL ANALYSIS:

• Status update: Search phase completed successfully
• Operation: Replace status from “in_progress” to “finished_successfully”

PHASE 5.5: CRITICAL DISTINCTION – SEARCH RESULTS vs PRODUCT LOOKUPS

Two Distinct Information Retrieval Systems

The ChatGPT SSE response reveals two fundamentally different systems operating in parallel:

SEARCH RESULTS SYSTEM (Lines 15-97)

Purpose: Gather informational content, reviews, and editorial recommendations
Data Source: General web search via Sonicberry/web.run tool
Content Type: Articles, blog posts, Reddit discussions, editorial content

Technical Characteristics:

{
  "type": "search_result_group",
  "domain": "marieclaire.co.uk", 
  "entries": [{
    "type": "search_result",
    "url": "https://marieclaire.co.uk/beauty/skincare/best-mineral-sunscreens-for-face",
    "title": "For Instant Protection and Unmatched Glow...", 
    "snippet": "This skincare guide from Marie Claire highlights...",
    "ref_id": null,
    "pub_date": 1754050495.0,
    "attribution": "www.marieclaire.co.uk"
  }]
}

Key Features:

• Domain grouping: Results organized by website
• Snippet extraction: Content summaries included
• Publication dates: Temporal relevance tracking
• Attribution: Source attribution for citation
• No pricing: Focus on information, not commerce

PRODUCT LOOKUP SYSTEM (Lines 193-235)

Purpose: Retrieve live commerce data for specific products
Data Source: E-commerce APIs and product databases
Content Type: Product specifications, pricing, ratings, merchant availability

Technical Characteristics:

{
  "query": "deprecated",
  "provider": "deprecated", 
  "id": "1029485709641249511",
  "title": "EltaMD UV Clear Broad-Spectrum SPF 46",
  "price": "$74.00",
  "merchants": "Mix Beauty Lab + others",
  "num_reviews": 763,
  "rating": 4.9,
  "product_lookup_data": {
    "request_query": "EltaMD UV Clear Broad-Spectrum SPF 46",
    "all_ids": {"p2": ["1029485709641249511"]},
    "metadata_sources": ["p2", "p1", "p3"]
  }
}

Key Features:

• Live pricing: Real-time price data from multiple sources
• Merchant tracking: Availability across different retailers
• Review aggregation: Rating scores and review counts
• Product IDs: Unique identifiers across platforms
• Image assets: Product photography URLs
• Structured data: Rich commerce-specific metadata

Critical Differences for SEO Professionals

Technical Implementation Insights

Search Results Processing:

1. Query sent to general web search engine
2. Results filtered and ranked by relevance
3. Snippets extracted for context
4. Domain grouping for organization
5. Attribution prepared for citation

Product Lookup Processing:

1. Product names extracted from search content
2. Queries sent to e-commerce APIs (p1, p2, p3 providers)
3. Live data retrieved (pricing, reviews, availability)
4. Product entities created with structured metadata
5. Merchant-specific information aggregated

SEO STRATEGY IMPLICATIONS:

• Content SEO: Focus on being included in search results system
• Product SEO: Optimize for product lookup system inclusion
• Dual optimization: Best results require presence in both systems
• Local commerce: Geographic targeting affects both systems differently

PHASE 6: CONTENT GENERATION INITIALIZATION (Lines 98-103)

Lines 98-103: Response Assembly Setup

RAW DATA:

99→event: delta
100→data: {"p": "", "o": "add", "v": {"message": {"id": "539a514a-46f4-439a-ad9a-86bb989119a5", "author": {"role": "assistant", "name": null, "metadata": {"sonicberry_model_id": "current_sonicberry_unpaid_oai", "real_author": "tool:web", "source": "sonic_tool"}}, "create_time": 1754307689.207317, "update_time": null, "content": {"content_type": "text", "parts": [""]}, "status": "in_progress", "end_turn": true, "weight": 1.0, "metadata": {"citations": [], "content_references": [], "search_turns_count": 1, "search_source": "composer_auto", "client_reported_search_source": "composer_auto", "request_id": "969dc9188d2fac9f-DFW", "message_type": "next", "model_slug": "gpt-4o", "default_model_slug": "auto", "parent_id": "3b417387-5632-4fb9-9f22-6aaf3b0d9b6a", "model_switcher_deny": []}, "recipient": "all", "channel": null}, "conversation_id": "68909c60-03e8-8003-812d-57045fb81d7d", "error": null}, "c": 4}   
101→
102→data: {"type": "message_marker", "conversation_id": "68909c60-03e8-8003-812d-57045fb81d7d", "message_id": "539a514a-46f4-439a-ad9a-86bb989119a5", "marker": "user_visible_token", "event": "first"}

TECHNICAL ANALYSIS:

• New message ID: 539a514a-46f4-439a-ad9a-86bb989119a5 – Response message identifier
• Create time: 1754307689.207317 – 3.83 seconds after search initiated
• Empty content parts: Starting with blank content for incremental generation
• User visible token: First token that will be displayed to user
• End turn: true – This will be the final response message

SEO INSIGHTS: 3.8-second delay from search to response generation shows real-time processing pipeline

PHASE 7: INCREMENTAL CONTENT GENERATION (Lines 104-296)

Lines 104-108: Opening Response Generation

RAW DATA:

104→event: delta
105→data: {"o": "patch", "v": [{"p": "/message/content/parts/0", "o": "append", "v": "If you're dealing with"}, {"p": "/message/metadata/safe_urls", "o": "add", "v": []}]}  
106→
107→event: delta
108→data: {"p": "/message/content/parts/0", "o": "append", "v": " oily or acne-prone skin and"}

TECHNICAL ANALYSIS:

• Patch operation: Multiple simultaneous updates to content and metadata
• Content generation: “If you’re dealing with oily or acne-prone skin and”
• Safe URLs: Initializing URL safety tracking list
• Incremental text: Building response word by word

Lines 109-123: Product Introduction & First Citation

RAW DATA:

111→data: {"v": " shopping around Toronto, here are some top-rated, oil-free sunscreens that offer strong UV protection and a matte or weightless finish.\n\n---\n\n## 🌟 Expert Recommendations for Oily Skin in Canada\n\nAccording to recent dermatologist-vetted guides and Canadian sources, these sunscreens stand out:\n\n- **La Roche-Posay Anthelios Clear Skin (Dry-Touch SPF 60/50)**  \n  A favorite for oily/acne-prone skin, it absorbs excess oil with a dry-touch finish and offers high broad-spectrum protection "}      
...
123→data: {"p": "", "o": "patch", "v": [{"p": "/message/content/parts/0", "o": "append", "v": " offers high broad-spectrum protection \ue200cite\ue202turn0search0\ue202turn0search2\ue202"}, {"p": "/message/metadata/content_references", "o": "append", "v": [{"matched_text": "\ue200cite\ue202turn0search0\ue202turn0search2\ue202", "start_idx": 538, "end_idx": 570, "safe_urls": [], "refs": [], "alt": null, "prompt_text": null, "type": "hidden", "invalid": true}]}]}

TECHNICAL ANALYSIS:

• Content structure: Using markdown formatting with headers and bullet points
• Location targeting: “shopping around Toronto” – addressing user’s geographic context
• Citation system: \ue200cite\ue202turn0search0\ue202turn0search2\ue202 – Unicode private use area encoding for citations
• Content references: Tracking citation metadata with start/end indices
• Product focus: La Roche-Posay as first recommendation

Lines 124-158: Citation System Development

TECHNICAL ANALYSIS: Lines 124-158 show the sophisticated citation management system:

• Citation encoding: Uses Unicode private use characters to embed citation references
• Reference resolution: System converts citation codes to actual URLs and sources
• Grouped citations: Multiple sources combined into single citation
• Attribution tracking: Each piece of information tracked to source articles
• URL moderation: Safety checking for all referenced URLs

Key Citation Examples:

• turn0search0 → fluidmediallc.com article
• turn0search2 → straight.com Canadian sunscreen guide
• turn0news13 → Real Simple dermatologist guide

Lines 159-193: Advanced Product Entity Integration

TECHNICAL ANALYSIS: This section marks the transition from search results to product lookups. The system now shifts from citing editorial content to retrieving live e-commerce data.

System Transition Point:
Lines 159-193 represent where ChatGPT moves from informational search to commercial product lookup:

1. Search citations end: No more turn0search references
2. Product queries begin: System starts turn0product lookups
3. E-commerce APIs activated: Live pricing and availability data
4. Structured product data: Rich commerce metadata integration

Lines 193-207: Product Lookup System Activation

RAW DATA: Lines 193-207 show advanced product entity system:

193→data: {"v": [{"p": "/message/content/parts/0", "o": "append", "v": "\ue200products\ue202{\"selections\":[[\"turn0product3\",\"EltaMD UV Clear Broad-Spectrum SPF 46\"],[\"turn0product7\",\"Supergoop! Mineral Mattescreen SPF 40\"],[\"turn0product5\",\"Ultra Violette Lean Screen SPF 50+ Mattifying Mineral\"]],\"tags\":[\"Derm-favourite niacinamide oil-control\",\"Invisible mineral mattifier\",\"Mattifying mineral SPF made for shine-control\"]}\ue201"

TECHNICAL ANALYSIS:

Product Lookup Architecture:

• Product selection: Three specific products identified with unique IDs
• Multi-provider system: Data aggregated from p1, p2, p3 e-commerce sources
• Live data integration: Real-time pricing, ratings, and availability
• Marketing intelligence: AI-generated product benefit tags
• Merchant aggregation: Multi-retailer availability tracking

Product Query Process:

1. Product name extraction: System identifies specific product names from search content
2. API queries dispatched: Parallel requests to multiple e-commerce providers
3. Data normalization: Pricing, ratings, and availability standardized
4. Merchant resolution: Retailer-specific information aggregated
5. Entity creation: Structured product objects with full metadata

Real-time Commerce Integration:
Unlike search results which may be cached, product lookups provide live commerce data:

• Pricing updated in real-time from merchant APIs
• Inventory status reflects current availability
• Review scores aggregate across platforms
• Merchant-specific shipping and pricing

Lines 208-235: Individual Product Entity Deep Dive

TECHNICAL ANALYSIS: Each product gets comprehensive entity treatment showing the sophistication of the product lookup system:

EltaMD UV Clear SPF 46 Analysis (Lines 208-215):

{
  "id": "1029485709641249511",
  "title": "EltaMD UV Clear Broad-Spectrum SPF 46", 
  "price": "$74.00",
  "rating": 4.9,
  "num_reviews": 763,
  "merchants": "Mix Beauty Lab + others",
  "product_lookup_data": {
    "metadata_sources": ["p2", "p1", "p3"],
    "variant_sources": ["p2"]
  }
}

Commerce Intelligence Features:

• Multi-source validation: Data verified across p1, p2, p3 providers
• Review aggregation: 763 reviews aggregated from multiple platforms
• Merchant diversity: “Mix Beauty Lab + others” indicates multi-retailer availability
• Price monitoring: Live pricing from primary merchant source
• Image assets: Product photography URLs for visual display

Supergoop! Mineral Mattescreen SPF 40 (Lines 216-225):

• Higher review volume: 3,300 reviews vs 763 for EltaMD
• Different merchant ecosystem: “Sephora + others” vs specialty retailers
• Price positioning: $54.00 mid-range positioning
• Rating variance: 4.4/5 vs 4.9/5 showing different market reception

Ultra Violette Lean Screen SPF 50+ (Lines 226-235):

• Emerging product: Only 89 reviews indicating newer/niche product
• Canadian focus: “Ultra Violette CA + others” emphasizing local availability
• Price advantage: $48.50 lowest price point
• Geographic optimization: CA-specific merchant prioritization

Lines 236-265: Product Comparison Intelligence

TECHNICAL ANALYSIS: The system generates structured comparison data combining both search results and product lookup information:

Hybrid Data Integration:
The comparison table demonstrates how ChatGPT combines:

1. Editorial insights (from search results): “High-oil, acne-prone skin”
2. Product specifications (from product lookups): SPF levels, pricing
3. Geographic intelligence (location-aware): “Shoppers Drug Mart, Well.ca”
4. User experience data (aggregated reviews): Finish types and skin compatibility

Intelligent Synthesis:

| La Roche-Posay Anthelios Clear   | Dry-touch matte| 50/60| High-oil, acne-prone skin     | Shoppers Drug Mart, Well.ca       |
| EltaMD UV Clear                  | Lightweight, matte | 46  | Sensitive + acne-prone skin   | Well.ca, online specialty retailers |

This shows the AI’s ability to:

• Synthesize data sources: Combine editorial opinions with commerce data
• Geographic targeting: Provide Toronto-specific retailer recommendations
• User segmentation: Match products to specific skin concerns
• Price-performance analysis: Balance cost with effectiveness ratings

Lines 266-288: Practical Application Guidance

TECHNICAL ANALYSIS: Final content generation demonstrates actionable intelligence synthesis:

Multi-source Knowledge Integration:

• Dermatological guidance: “two finger-lengths” application advice
• Technical specifications: “silica, perlite, or gel textures” ingredient guidance
• User experience optimization: “fully absorb before makeup” practical tips
• Product-specific recommendations: Formulation guidance based on product lookup data

Lines 289-296: Response Finalization & System Completion

RAW DATA:

288→data: {"p": "", "o": "patch", "v": [{"p": "/message/content/parts/0", "o": "append", "v": "erin Oil Control**, **Kinesys**, or **ATTITUDE** oil-free versions.\n\nThese options should be widely available across major retailers in Toronto (Shoppers Drug Mart, well-known beauty sites, and online Canadian marketplaces).\n\nNeed help finding where to buy a specific product nearby, or choosing between formulations? I'd be happy to help!"}, {"p": "/message/status", "o": "replace", "v": "finished_successfully"}]}
290→data: {"type": "message_stream_complete", "conversation_id": "68909c60-03e8-8003-812d-57045fb81d7d"}

TECHNICAL ANALYSIS:

• Final recommendations: Canadian brand alternatives (showing search result influence)
• Local availability: Toronto-specific retailer mentions (geographic intelligence)
• Call to action: Invitation for follow-up questions
• Stream completion: Final protocol closure
• Status finalization: Message marked as successfully completed

Dual System Success Indicators:

• Search results impact: Canadian brand recommendations from editorial content
• Product lookup impact: Specific retailer availability from commerce APIs
• Geographic precision: Toronto-specific recommendations combining both systems
• User engagement: Follow-up invitation indicating AI confidence in response quality

COMPREHENSIVE TECHNICAL INSIGHTS

SEO & AI Search Implications

1. Dual-system architecture: Search + Product lookup creates comprehensive responses
2. Real-time commerce integration: Live pricing affects search ranking and recommendations
3. Location-aware processing: Geographic context influences both content and commerce results
4. Multi-source validation: Editorial content and commerce data cross-validate recommendations
5. Attribution transparency: Complete source tracking across both information systems
6. Progressive enhancement: Complex responses built through incremental data integration

Technical Architecture Insights

1. Delta encoding efficiency: Incremental updates reduce bandwidth for complex responses
2. Parallel processing: Search, product lookup, and content generation operate simultaneously
3. Citation management: Sophisticated source tracking spans both editorial and commerce sources
4. Content moderation: URL safety checking and content validation across all data sources
5. Infrastructure optimization: Prewarmed servers handle multi-system integration seamlessly
6. Entity resolution: Product lookup system provides structured commerce data integration

Advanced Features Observed

1. Sonic classifier: 68.2% confidence threshold triggers both search and product lookup systems
2. Typo tolerance: System handled “Oily Screen” → “Oily Skin” across both information systems
3. Contextual understanding: Recognized product purchase intent with location for dual-system activation
4. Multi-modal response: Text, tables, product cards, and citations from integrated data sources
5. Progressive enhancement: Building complex response through coordinated system integration
6. Quality control: Multiple validation layers across search results and product data

Commerce vs Content Intelligence

Search Results System Strengths:

• Editorial authority and expert opinions
• Comprehensive product category coverage
• User experience insights and reviews
• Geographic and cultural context
• Trending information and current discussions

Product Lookup System Strengths:

• Real-time pricing and availability
• Merchant-specific information
• Structured product specifications
• Review aggregation across platforms
• Inventory and shipping intelligence

Combined System Power:
The integration of both systems creates responses that are:

• Contextually rich (editorial insights) + Commercially actionable (purchase data)
• Authoritative (expert sources) + Current (live commerce data)
• Comprehensive (category coverage) + Specific (individual products)
• Educational (how-to guidance) + Transactional (where to buy)

This analysis reveals ChatGPT’s sophisticated dual-system approach to commercial search queries, seamlessly combining real-time web search with live e-commerce data integration to provide comprehensive, actionable responses that serve both research and purchase intent.

Earlier, I used to use Add-Ons like GPT for Sheets, but slowly they all became paid.

So I figured a free alternative for my needs. I’m sharing it here so you all can use it as well.

Step 1: Generate OpenAI API Key

Go to this link : https://platform.openai.com/api-keys

Click on “Create new secret key”

Give it a name so you can identify where the key was used.

Copy the key as you will need it later.

Step 2: Create an App Script

This will bring you to a view like this.

Remove everything from the code section, and Paste this code into the App Script

Apps Script Code

function GPT(prompt, systemMessage, temperature, model) {
  const apiKey = 'YOUR_OPENAI_API_KEY'; // Replace this securely
  const url = 'https://api.openai.com/v1/chat/completions';

  // Set defaults if values are not provided
  systemMessage = systemMessage || "You are a helpful assistant.";
  temperature = typeof temperature === 'number' ? temperature : 0.7;
  model = model || "gpt-4o";

  if (!prompt || prompt.trim() === "") {
    return " Please provide a prompt.";
  }

  const payload = {
    model: model,
    messages: [
      { role: "system", content: systemMessage },
      { role: "user", content: prompt }
    ],
    temperature: temperature
  };

  const options = {
    method: 'post',
    contentType: 'application/json',
    headers: {
      Authorization: 'Bearer ' + apiKey
    },
    payload: JSON.stringify(payload),
    muteHttpExceptions: true
  };

  try {
    const response = UrlFetchApp.fetch(url, options);
    const json = JSON.parse(response.getContentText());

    if (!json.choices || !json.choices[0]) {
      return " No response from OpenAI.";
    }

    return json.choices[0].message.content.trim();
  } catch (e) {
    return " Error: " + e.message;
  }
}

Replace YOUR_OPENAI_API_KEY with your own API Key that you generated.

Click on Save Button.

Step 3: Make an OpenAI call from Google Sheet.

Now go to your Google Sheet, and you can make an API call.

It should return a response like this.

To make the experience a little bit better, I’ve baked in some customization like model selection, system prompt, temperature and error handling. Here are some examples of how you can use them all.

You’re All Set!

With just a few lines of code, you’ve unlocked the power of GPT right inside Google Sheets — totally free.

]]> https://www.mihirnaik.com/how-to-make-openai-api-calls-from-google-sheets/feed/ 0 https://www.mihirnaik.com/my-seo-strategy-framework/ https://www.mihirnaik.com/my-seo-strategy-framework/#comments Sun, 31 Oct 2021 17:04:11 +0000 https://www.mihirnaik.com/?p=28454