ChatGPT crossed 900 million weekly active users in February 2026 (TechCrunch, Feb 2026). Buyers are increasingly starting there instead of in Google, asking "who should I use for X?" and trusting whatever name comes back. For B2B brands, that means the way an LLM describes your company is often the first impression a buyer gets of you, and you have very little control over what it says.
The way an LLM describes your company is often the first impression a buyer gets, and you have very little control over what it says.
Most B2B teams have already seen the gap. ChatGPT misattributes a product, Perplexity lists a competitor's pricing under their name, Claude skips them entirely. There's no settled playbook for fixing it, and most "GEO guides" treat the problem like SEO with new keywords. That framing misses the point. The pipeline behind a generated answer doesn't work the way PageRank-style ranking does, so applying the same tactics keeps you misattributed.
What follows: what's changing in AI brand visibility, what to measure, and what to do about it this week. Every claim has a source.
What is generative engine optimization, really?
Generative engine optimization (GEO) is the practice of getting your brand correctly cited, recommended, and described inside AI-generated answers instead of inside ranked search results. The surfaces that matter are ChatGPT, Claude, Perplexity, and Google AI Overviews. Academic work by Aggarwal et al. (Princeton, KDD '24) showed GEO tactics deliver a 30-40% relative lift on the Position-Adjusted Word Count metric, with citation-format passages driving the largest gains.
AEO (answer engine optimization) is the older, snippet-focused variant. LLMO (LLM optimization) is the engineer's term. AIO (AI optimization) is the version that shows up in vendor decks. Wikipedia treats the three as synonyms, so GEO is the umbrella we use throughout this post.
It is not "SEO with a new acronym"
SEO optimizes for a ranked list of URLs. GEO optimizes for a sentence inside a generated answer. The pipeline behind that sentence (retrieval-augmented generation, structured-data ingestion, citation indexing) has different mechanics from PageRank-style link analysis, which means a passage ranking #3 on Google might not be cited at all by Claude, and vice versa.
Most top-ranking posts about GEO frame this as a content problem (the implied fix being "write better citation passages"). That captures half of it. Equally important is the infrastructure layer underneath: bot-detectable URLs, machine-readable brand data, and something between the AI crawler and your origin so the bot doesn't end up guessing at what's behind the JavaScript wall.
How fast is this shift moving?
AI-referred sessions grew 527% year-over-year in the first five months of 2025 (Previsible 2025 AI Traffic Report, via Frase), and ChatGPT alone reached 900M weekly active users by February 2026 (TechCrunch). The underlying point isn't the absolute number, which everyone has already seen, but the slope: AI search went from rounding-error traffic to a tracked channel inside eighteen months.
Google's AI Overviews (the answer box that pre-empts the blue links) fired on 6.49% of US desktop searches in January 2025, peaked near 25% in July, and settled at 15.69% by November (Semrush AI Overviews study, 2025), while Gartner forecasts a 25% drop in traditional search volume by the end of 2026 (Gartner press release, Feb 2024). Two different methodologies, same direction.
Almost all of that traffic is one tool:
Chart data
| Assistant | Referral share |
|---|---|
| ChatGPT | 89.10% |
| Microsoft Copilot | 3.20% |
| Perplexity | 3.10% |
| Google Gemini | 2.40% |
| Claude | 1.35% |
| Grok | 0.55% |
| DeepSeek | 0.20% |
ChatGPT alone takes 89.1% of AI referral traffic. The rest of the field is a long tail (Goodie / Similarweb, 2025). The intent-action gap on top of that is the awkward part: 40.6% of marketers cite updating SEO for AI search as a top trend they're actively tackling this year (HubSpot 2026 State of Marketing Report), but only 16% of brands systematically track AI search performance (McKinsey, September 2025 CMO survey). Most of the marketers who know they should be doing something about this still aren't measuring whether it works.
What changed between SEO and GEO?
SEO optimizes for ranking on a SERP, while GEO optimizes for citation inside a generated answer. The artefact you produce is no longer a page, it is a passage an LLM can lift verbatim. Only 33% of users who see an AI-generated answer say they always or often click through to a source. Another 37% click only sometimes (Reuters Institute, 2025). For the rest, what the model says about you is all they get.
The two playbooks, side by side
Old playbook: backlinks, keyword targeting, on-page schema, content velocity. The output is a page that ranks. New playbook: structured brand intelligence, bot interception at the edge, citation-format passages, ongoing perception tracking across models. The output is a sentence an LLM is willing to attribute to you.
| SEO era | GEO era | |
|---|---|---|
| What you produce | A page that ranks | A passage an LLM can cite |
| What you measure | Rank position | Citation rate, sentiment, model coverage |
| When you improve | One number moves up or down | Three dimensions move independently |
The conversion numbers are the kicker. In a Seer Interactive case study tracking one B2B client across seven months (October 2024 through April 2025), LLM referral traffic converted at much higher rates than Google organic: ChatGPT 15.9%, Perplexity 10.5%, Claude 5%, and Gemini 3%, versus 1.76% for Google organic (Seer Interactive, June 2025). N=1 site is a small sample, but the directional gap (LLM referrals converting at single-to-double-digit multiples of Google organic) shows up across multiple recent case studies. By the time someone hits your site from an AI prompt, they're far down the funnel.
What does the new measurement stack look like?
You can't optimize what you can't see, and what's visible in GA4 is only a small slice of what's happening. A working GEO measurement stack has three layers: bot traffic (who's crawling), brand perception (what the models say about you), and conversion (which AI tools send buyers). Only 16% of brands systematically track any of this today (McKinsey, September 2025 CMO survey), which is partly why the early movers can still pull ahead.
Layer 1: bot traffic at the edge
Server-side bot detection catches GPTBot, ClaudeBot, PerplexityBot, and Google-Extended before they ever touch your origin. You see fetch frequency, which URLs they pull, and whether they come back. Most analytics suites ignore non-human traffic. For GEO that is exactly the traffic you want to study.
Layer 2: brand perception across models
Brand-tracker prompts run on a schedule, fired at multiple models, across personas and regions. The output is a longitudinal record of what ChatGPT, Claude, Perplexity, and Gemini say about your company. Sentiment, accuracy, and which competitors get named alongside you. This is the layer your CMO will care about.
Layer 3: AI-referred conversions in GA4
Segment GA4 sessions by referrer to surface chat.openai.com, perplexity.ai, claude.ai, and gemini.google.com. Tag them as a distinct channel. Once you have a few weeks of data you can compare close-rates against organic, paid, and direct.
What is the wince test and how do you run it on your own brand?
The wince test is a five-question protocol you can run on your own brand in twenty minutes. Ask each of GPT-4.1, Claude Sonnet, and Perplexity Sonar Pro the same five questions about your company. The gap between what you sell and what AI says you sell is your first GEO backlog. It is also where most founders wince. Top-tier frontier models on HHEM-2.3 sit between 1.8% and 6% hallucination on standardized summarization, while full-scale reasoning models like GPT-5, Claude Sonnet 4.5, and Gemini 3 Pro sit higher in the 8 to 15% range (Vectara HHEM-2.3 leaderboard), and brand-specific questions are harder still because the source-of-truth lives on your site, not in the prompt.
The five questions
- Overview. "What does <company> do?" Check that the description matches your actual positioning.
- Products. "What products does <company> sell?" Watch for missing SKUs, ghost features, or invented bundles.
- Competitors. "Who are <company>'s main competitors?" Note who is named, who is missing, and who is wrongly grouped.
- Recommendation. "Should I use <company> for <your core use case>?" Look at the reasoning, not just the verdict.
- Reputation. "What do people say about <company>?" Surface what reviews, Reddit threads, and old blog posts have fed the model.
Our team ran the wince test on cloudweld.ai (the company building Ooky) before writing this post. GPT-4.1 described the parent brand accurately, Claude got the product category right but invented a pricing tier that has never existed, and Perplexity attributed a competitor's case study to us. Three models, three different failure modes, and we did not need any customer data to find them. The deeper walk-through lives in the five questions to ask ChatGPT about your brand.
Running it manually takes about twenty minutes for one company at one moment in time. Running it every week, across every major model, against the dozen prompts a buyer might type, is more work than any founder has time for. That is the piece Ooky's free tier handles for you: it schedules the prompts, captures the answers across models, and flags the diff against your brand intelligence as it updates. See what the free tier covers if you want the wince test to keep running after the first manual pass.
How do you fix AI misattribution?
More content alone doesn't help, because the problem isn't that the model can't find words about you. The problem is that the model is filling gaps by inferring. What helps is putting a small machine-readable description of your brand between the AI bot and your origin, so the model has something concrete to retrieve when the bot shows up. Models infer when they can't retrieve. Even Vectara's HHEM-2.3 leaderboard, which measures summarization faithfulness on prompts that include the source text, shows top-tier models between 1.8% and 6% and full-scale frontier reasoning models (GPT-5, Claude Sonnet 4.5, Gemini 3 Pro) clustering in the 8 to 15% band (Vectara HHEM-2.3, 2026). Brand questions are harder because the source-of-truth lives on your site, not in the prompt. Give models something authoritative to retrieve and the inference rate drops.
Why this has to happen at the edge
Timing is the part most teams miss. A bot lands, fetches a page, and decides what to remember in the same request. If your truth lives behind JavaScript or three clicks deep, the model never reaches it and infers instead. The fix has to meet the bot at the door with a clean, structured answer before it ever touches your origin. Ooky runs that interception for GPTBot, ClaudeBot, PerplexityBot, Google-Extended, and Bytespider, while human traffic is untouched.
Why "do not infer" exists
Some fields should never be guessed at: pricing, customer names, regulated claims, anything where a wrong sentence costs you a deal or a regulator visit. Ooky's brand intelligence flags those fields so the retrieving model knows to leave the answer out rather than improvise. We unpack why this directive matters in the next post.
Content-only GEO advice falls short because content on your site doesn't reach a model that hasn't crawled lately. A model without your current pricing page will happily paraphrase from a 2023 Reddit thread instead. The fix isn't writing a better Reddit thread. It's making sure the model finds something authoritative at retrieval time.
FAQ
Is GEO replacing SEO?
No. Search has not disappeared, but its share is shrinking. Gartner forecasts a 25% drop in traditional search volume by the end of 2026 (Gartner press release, Feb 2024). By October 2025, informational queries accounted for 57.1% of AI Overview triggers, commercial 18.57%, and transactional 13.94% (Semrush AI Overviews study). Treat GEO as the second lane, not the replacement.
How do I get my website cited by ChatGPT?
Two prerequisites. Be crawlable by GPTBot, PerplexityBot, ClaudeBot, or Google-Extended. And host machine-readable, factually clean brand information at predictable URLs. Self-contained citation passages of 40 to 60 words containing a claim plus a source deliver a 30-40% relative lift on the Position-Adjusted Word Count metric (Aggarwal et al., KDD '24). The mechanic rewards specificity and source attribution.
What is the difference between GEO, AEO, and LLMO?
They're variants of the same job. GEO (generative engine optimization) is the umbrella term and the one we use. AEO (answer engine optimization) is older and narrower, focused on featured snippets and direct answers. LLMO (LLM optimization) is the engineer-flavoured label. Wikipedia treats the three as synonyms; vendors use whichever sells.
How do I track AI search referrals?
Server-side bot logs, brand-tracker prompts fired at the major models on a schedule, and GA4 organic-AI referrer segmentation are the three signals. The data isn't the hard part; the hard part is that only 16% of brands look at any of it (McKinsey, September 2025 CMO survey), so even getting one signal in place puts you ahead of most peers.
How do you start the GEO work this week?
Run the wince test on your own brand this week. Ask the same five questions of GPT-4.1, Claude Sonnet, and Perplexity Sonar Pro, and note the gap between what you sell and what they say you sell. That gap is your starting backlog. The intent-action gap is real: 40.6% of marketers are updating SEO for AI search this year (HubSpot 2026), but only 16% are tracking it (McKinsey 2025). Most of your peers will get to this work the same week (or quarter, or year), whether they wanted to or not.
Put structured brand intelligence in front of the bots
Ooky intercepts AI crawlers before they hit your origin and serves them the truth. The free
tier covers bot-event capture, perception tracking across every major model, and do_not_infer directives on sensitive fields. Start there.
Sources
- TechCrunch. "ChatGPT reaches 900M weekly active users." February 27, 2026. Retrieved 2026-06-07. https://techcrunch.com/2026/02/27/chatgpt-reaches-900m-weekly-active-users/
- Gartner. "Gartner Predicts Search Engine Volume Will Drop 25% by 2026, Due to AI Chatbots and Other Virtual Agents." February 19, 2024. Retrieved 2026-06-07. https://www.gartner.com/en/newsroom/press-releases/2024-02-19-gartner-predicts-search-engine-volume-will-drop-25-percent-by-2026-due-to-ai-chatbots-and-other-virtual-agents
- Semrush. "AI Overviews study (10M keyword sample)." 2025. Retrieved 2026-06-07. https://www.semrush.com/blog/semrush-ai-overviews-study/
- Previsible (via Frase). "2025 AI Traffic Report: what is generative engine optimization (GEO)." 2025. Retrieved 2026-06-07. https://www.frase.io/blog/what-is-generative-engine-optimization-geo
- Goodie / Similarweb. "AI search market share report." 2025. Retrieved 2026-06-07. https://higoodie.com/blog/ai-search-market-share-report/
- Seer Interactive. "Case Study: 6 Learnings - How Traffic from ChatGPT Converts." June 3, 2025. Retrieved 2026-06-07. https://www.seerinteractive.com/insights/case-study-6-learnings-about-how-traffic-from-chatgpt-converts
- Aggarwal, P. et al. "GEO: Generative Engine Optimization." arXiv / ACM KDD '24. August 2024. Retrieved 2026-06-07. https://arxiv.org/abs/2311.09735
- Vectara. "Hallucination Evaluation Model (HHEM-2.3) leaderboard." Updated May 2026. Retrieved 2026-06-07. https://github.com/vectara/hallucination-leaderboard
- McKinsey & Company. "New front door to the internet: Winning in the age of AI search." September 2025 CMO survey (~30 Fortune 500 consumer-brand CMOs). Retrieved 2026-06-07. https://www.mckinsey.com/capabilities/growth-marketing-and-sales/our-insights/new-front-door-to-the-internet-winning-in-the-age-of-ai-search
- HubSpot. "2026 State of Marketing Report." 2026. Retrieved 2026-06-07. https://blog.hubspot.com/marketing/hubspot-blog-marketing-industry-trends-report
- Reuters Institute for the Study of Journalism. "Generative AI and News Report 2025." 2025. Retrieved 2026-06-07. https://reutersinstitute.politics.ox.ac.uk/generative-ai-and-news-report-2025-how-people-think-about-ais-role-journalism-and-society