Why this changed
When Google shows an AI summary, people click through far less. Pew watched what 900 US adults actually did across 68,879 searches in March 2025: with an AI summary on the page, users clicked a result 8% of the time, against 15% when there was no summary. They ended the session entirely 26% of the time, against 16%.
That is the whole shift in one number. Being listed is worth less than it was; being quoted is worth more. Both still start from the same place — a page a machine can read.
Source: Pew Research Center, July 2025 ↗What actually makes a page quotable
The one peer-reviewed study on this tested nine tactics across 10,000 queries. What raised visibility: quoting a source, citing statistics with attribution, naming your sources, and writing fluent plain language. What did nothing at all: keyword stuffing — the single clear negative result in the paper.
Read those numbers as direction, not gospel. The study's main engine was simulated on an older model and the work is from 2023–24; the authors say themselves the methods will need to adapt as engines change.
Google's own position is blunter: there are no extra requirements to appear in AI Overviews, no special AI text files, and no special schema you must add. The one thing it does ask for is that your structured data matches the words visible on the page.
Source: Aggarwal et al., GEO (KDD 2024) ↗Ranking is no longer the same as being cited
Ahrefs studied 863,000 result pages and four million AI Overview URLs in March 2026: only about 38% of AI Overview citations came from pages ranking in the top ten, down from roughly 76% in July 2025. About 37% did not rank in the top hundred at all.
Ahrefs notes part of that drop is their own improved parsing. Even discounted, the direction is clear — a page can be quoted without ranking, and can rank without being quoted.
Source: Ahrefs, March 2026 ↗The structure we build in
Answer first, in under about fifty words, before any preamble. Question-shaped headings, because that is what a query looks like. Real tables rather than styled boxes — an extractor reads the document structure, and a grid of divs is invisible as a table. Short, declarative sentences of fact. A visible last-updated date in real markup.
Honest note: these are craft consensus and mechanically sound, not laboratory-proven. We build them in because they cost nothing and the mechanism is sound — not because someone measured a conversion lift we can quote at you.
llms.txt — we will not sell you this
llms.txt is a proposed file telling AI models how to read your site. Google has said publicly it does not support it and is not planning to, and compared it to the old meta keywords tag. No major model vendor has committed to reading it in production.
It takes ten minutes and harms nothing, so we are happy to add one. But there is currently no evidence it produces a single citation, and anyone selling llms.txt as an AI-visibility service is selling a hypothesis.
Source: Search Engine Journal ↗IndexNow — real, and narrower than it sounds
IndexNow pushes a notification the moment a page changes. Bing, Yandex, Naver, Seznam, Yep and Amazon honour it. Google does not — Google still needs your sitemap and Search Console.
It matters anyway for one specific reason: ChatGPT's web search runs on Bing's index. Getting into Bing quickly is the ChatGPT and Copilot pipeline.
Source: IndexNow FAQ ↗For an Indian shop, in order of what actually moves
1. Your Google Business Profile. Google ranks local results on relevance, distance and prominence — for a shop the profile outranks the website. Hours, category, photos and review volume are the levers you actually control.
2. Reviews, and the same name, address and phone number everywhere. That consistency is what an AI cross-checks before it repeats a fact about you.
3. Your city and locality in the page heading and title — one real page per area you serve, not forty near-identical generated ones.
4. LocalBusiness structured data that matches what the page visibly says. This is Google's one stated requirement.
5. llms.txt, last and optional.
Source: Google local ranking factors ↗Three things people get wrong
FAQ schema no longer earns a Google rich result — those were restricted in 2023 and removed entirely on 7 May 2026. FAQ content still helps a machine extract answers; the schema simply stopped buying you a listing. Anyone still selling FAQ markup for rich results is quoting a dead feature.
AEO is not a separate product with separate deliverables. It is a formatting layer on genuinely helpful content — Google's stated position, and ours.
Structure into a void does nothing. A page nobody has indexed cannot be cited, because models draw on indexed, trusted sources. Get indexed first, then optimise for being quoted.
Source: Google: FAQPage deprecation ↗The numbers we refuse to quote
Every agency blog repeats the same India local-search statistics — that 46% of searches have local intent, that 78% of local mobile searches end in an offline purchase, that near-me searches in India rose 250%. We traced none of them to a primary source; the 250% figure is a global number about one specific phrase, misattributed to India.
We have left them out. A page about being trustworthy to machines is a strange place to start inventing figures.