If you’re still arguing that AI search is something to watch rather than act on, the data makes that difficult to grasp. Currently, 30% of consumers use AI in product research. Last year, that figure was 12%. And almost 40% of Google results now include AI Overview, which means people are reading AI-generated answers without exception.
How many people actually use AI search?
Put those two things together and the share of search journeys involving AI is already close to 60%. Brainlabs projects to reach 80% within twelve months, based on current growth rates. That changes the way you think about almost everything in organic search.
Brainlabs survey data divides consumers into three groups. Traditionalists (about 69%) use only established search engines. Augmenters (about 30%) use both Google and AI platforms in different areas in a single research trip. The opposition (less than 1%) use only AI platforms. Augmenters are where the growth is – they haven’t abandoned Google, they’ve added AI as a research layer for complex queries.
Why AI SEO is not just a reinvention of traditional SEO
Honest answer: it’s similar enough right now that you don’t need to redo your roadmap, but different enough that the gap is growing faster than most teams are prepared for. There is no single search AI – there are at least four major platforms with different models, different guidelines, and different relationships with Google’s index.
That 15% Gemini figure is what we need to stop at. Gemini is a Google product, yet it only cites the top 10 Google results pages 15% of the time. An SEO strategy built entirely on top 10 rankings does very little for Gemini or ChatGPT. And the overlap is constant — the AI overview used to have 76% overlap with Google’s top 10. By 2026, it has halved.
How do LLMs actually decide what to say
Understanding why an obscure page ends up being mentioned in a Gemini response while the main product content is ignored requires understanding how LLMs generate responses. It’s really different from how Google ranks pages.
Bottom line: When Gemini receives information, it wonders whether it needs external data or can rely on training data alone. For most product research questions, it requires external data.
Question fan-out: The LLM generates a number of related sub-questions from the original material. For complex information there may be tens of these working simultaneously.
In-depth index search: For each sub-question, LLM searches the Google index of the top 100 results and above. Not just the top 10 – this is the first big difference to traditional SEO.
Content selection: LLM looks for specific answer blocks, topics related to sub-questions, strong statistical data, and new signals such as a recent “last updated” date.
Comparison and validation: Sources with low consensus compared to high authority references are filtered out. A page that answers a small question directly can compete with top pages that keep their answer to editorial prose.
What LLMs look for when choosing citations
- Direct answers: Paragraphs that lead quickly through the answer, not the planning structure
- Heading-level relevance: H2s and H3s closely match potential sub-questions
- Hard statistical data: Specific numbers, dates, percentages that the model can extract and quote
- New signals: A clearly visible “last updated” date, within the last 30–60 days for sensitive topics
- Clean, organized HTML: Tables, lists, and semantic markup LLM can be easily analyzed
The three optimization methods produced the most consistent results. First, read your fan-out question – the little questions that LLMs do know how to know, and FAQ sections built around them drive measurable citation increases. Second, use embedding analysis to measure how closely your content matches what’s being said; this produced an average 140% increase in AI citations across all Brainlabs client trials. Third, treat content freshness as a technical requirement — monthly updates of the most in-demand pages are a minimum for time-sensitive topics.
Measuring AI search performance is difficult. Here is the reason.
No first-person data from any major AI platform is available at scale. You can’t see actual user queries on Gemini. Without that data, everything is proxy and estimates, which introduce real uncertainty into any measurement framework.
The current standard approach: convert keyword data into potential orders, track them in one of 30-plus third-party AI tracking tools, and measure product visibility over time as a proxy for real-world performance. The problem is that data is noisier than it looks. Across all models, only 23% of quotes are still valid after 14 days. There is only about 45% agreement within the field as to which brand should be recommended first. And less than 5% of query sets show perfect consistency across all major platforms.
Effective measurement methods are now in effect
- Use 100 relevant pieces of information per category, four times each – the same reliability as using one piece of information 400 times, with rich category insights
- Track AI referral traffic in GA4 and perform correlation analysis against existing Google search demand data
- Use published AI search studies (SEOClarity has high-quality data) and apply reasonable adjustments to existing Google volume estimates
- Document your methodology – any AI traffic measurement will be questioned by leadership, so reasoning must be protected
Agent search: next shift
“We are still distributing [agentic search] … and I really expect in some of these places, [20]27 to be a critical turning point.
Sundar Pichai, CEO, Google
Everything so far sits within the current paradigm, where AI is a layer on top of search. The next shift is big. Photosi expects search to transform into an “agent manager” — where AI agents conduct research and make decisions on behalf of the consumer. In the emerging agent model, an AI agent searches in seconds, filters to a recommendation, and guides the transaction, all without the consumer leaving the AI platform. ChatGPT already has Instant Checkout. Google has a Universal Commerce Protocol.
Agents still need to do research, and they still need information that has to come from somewhere. The question is which sources of content and data will they trust enough to cite and act on.
Three ways to prepare an agent search now
- Reorder content for agent use. Agents look for clear conclusions, clear recommendations, accurate structured data, and clean HTML. Burying recommendations in the third category does not work at machine speed.
- Think of your data as the assets agents will want to connect to. B2B API access, consumer-facing authentication, and tiered licensing with AI platforms are all viable business models that don’t need to wait.
- Consider building rather than just eating. Brands with valuable professional content have the ingredients to become an agent platform themselves – with commercial rails for users to act on recommendations quickly.
Where will you start?
Start with an estimate. Set up AI transfer tracking on GA4, select two or three key categories, and start tracking 50 to 100 related information at low frequency. The data will not be complete, but it will be targeted.
Next, check your valuable content against the signals LLM is looking for: new dates, specific answer formatting, structured data, and subject-level relevance to potential sub-questions. Retargeting the most popular pages using these criteria will tell you if the changes are moving the quote values before scaling.
Then build a road of separation. The overlap between traditional SEO and AI SEO is decreasing. The teams that sit well in two years are the ones who are building those skills now. The window to get ahead, rather than rush, is shorter than it looks.
