How AI decides which businesses to recommend
When someone asks ChatGPT, Claude, Perplexity or Google's AI for "the best [your category] in [your town]," a handful of businesses get named, and the rest are invisible. That outcome isn't random. It's driven by six measurable signals. This is the framework we use to score them, and to move a business from invisible to recommended.
Visibility vs Readiness
Two different questions, two different scores. Most "SEO audits" conflate them, and that's why they don't predict whether AI actually recommends you.
Are you in the answer right now?
Whether the AI assistants actually name or cite you when a real customer asks. Mention-driven, measured live across up to seven engines. This is the outcome that wins customers.
Is your site built for AI?
How well your foundations support being recommended, the six signals below, scored 0–100. You can be highly ready yet still invisible: a well-built site the AIs don't yet know or trust.
The gap between them is the work: readiness is what you control; visibility is what it earns you.
That gap is not theoretical, it's the single biggest thing our own data shows. See what 2,354 real businesses prove below.
What AI weighs to decide
Citability
25% · highest weight- What
- How quotable and self-contained your content is, how easily an AI can lift a clear, factual passage straight into its answer.
- Why
- AI answers are assembled from quotable passages. Thin, vague or purely promotional copy gives the model nothing to use, so it reaches for a competitor who wrote a clear answer.
- Win it
- Answer real questions directly, lead with the conclusion, use headings, lists, statistics and specifics. Write the paragraph you'd want quoted.
Brand
20%- What
- Whether AI recognises you as a real, distinct entity and whether trusted third parties, Wikipedia, Wikidata, Reddit, YouTube, reviews, industry directories, reference you.
- Why
- AI recommends what others vouch for. A business with no third-party footprint reads as unknown and risky to name, however polished its own site. It is also, by a wide margin, the strongest predictor in our own data.
- Win it
- Earn citations off your own domain: authoritative directory listings, industry-association profiles, genuine reviews and press mentions, a consistent entity (same name, address, links) everywhere, and, where it's realistic for your size, a Wikipedia/Wikidata entry.
E-E-A-T
20%- What
- Experience, Expertise, Authoritativeness, Trust. Author credentials, a real About page, visible contact details, provenance and a privacy policy.
- Why
- Recommending a business is a trust decision. AI is built to avoid endorsing the untrustworthy, so it favours sites that prove who's behind them.
- Win it
- Name your people and their credentials, show how you know what you claim, make it obvious you're a real, contactable operation.
Technical
15%- What
- Whether AI crawlers can cleanly access and read you, HTTPS, sitemaps, canonical URLs, mobile, security headers, server-side rendering and Core Web Vitals.
- Why
- If the content an AI needs only appears after heavy JavaScript, or the crawler is blocked or timed out, none of the other signals matter: it never sees them.
- Win it
- Serve content in the initial HTML, keep it fast and crawlable, fix the plumbing. (See Foundation below, your platform decides how easily.)
Schema
10%- What
- Structured data (JSON-LD), Organization, LocalBusiness, Product, FAQ, that states what you are, where you operate and what you offer in machine-readable form.
- Why
- Schema removes guesswork. It hands the AI a clean, unambiguous fact sheet about your entity instead of making it infer everything from prose. On its own it's necessary, not sufficient, our own data shows why.
- Win it
- Add valid Organization + LocalBusiness JSON-LD, with sameAs links to your verified profiles. Cheap, fast, high-leverage.
Platform
10%- What
- Readiness for each AI surface, crawler access (robots.txt for GPTBot, Google-Extended, PerplexityBot, ClaudeBot), an llms.txt, and answer-shaped content tuned to how each engine retrieves.
- Why
- Each assistant sources answers differently. Quietly blocking AI crawlers, or never telling them what you do, guarantees you're absent from that surface.
- Win it
- Open the right crawlers, publish an llms.txt, and structure pages as direct answers to the questions customers actually ask the AI.
Your foundation
Underneath the six signals is your platform, and it decides how cheaply you can win them. A modern, controllable stack lets you fix schema, llms.txt, rendering and agent integration in hours. A closed legacy builder caps what you can do at all.
- Modern (Next.js, Astro, custom Cloudflare): full control of rendering, schema and llms.txt; easy to make agent- and GEO-ready.
- Workable (Shopify, Webflow, well-built WordPress): GEO-ready with deliberate setup; some ceilings on control.
- Legacy (Wix, Squarespace, page-builder WordPress): limited control over the exact signals AI relies on, often the single biggest constraint on visibility.
How we score it
An autonomous agent interrogates ChatGPT, Claude, Perplexity and Google's AI with the question your customers actually ask, then audits 30+ signals across the six dimensions above plus your foundation and agent-readiness, producing a live AI Visibility score and an AI Readiness score, with the specific fixes that close the gap. The deep audit adds account-gated signals, component versions and a human review. Whether you're calling it GEO, Answer Engine Optimisation, or just AI search optimisation, it's the same measurement, and this is the tool that runs it.
What is Generative Engine Optimisation (GEO)?
Generative Engine Optimisation (GEO) is the practice of structuring your business's online presence so that AI engines, ChatGPT, Claude, Perplexity, Google's AI Overviews and Gemini, find, trust, recommend and cite you when people ask them for help. Where SEO competes for a rank in a list of blue links, GEO competes to be the answer the AI gives. The term was introduced by researchers at Princeton, Georgia Tech, the Allen Institute for AI and IIT Delhi in 2023.
It matters now because search has moved. Google's AI Overviews already appear on roughly one in four searches, and on those queries organic click-through has fallen about 61% (Seer Interactive, 2025). When the AI answers in one box and names two or three businesses, ranking on page one of the old results is no longer the same as being found.
What is Answer Engine Optimisation (AEO)?
Answer Engine Optimisation (AEO) is the practice of structuring content so AI systems can lift it directly into an answer: a clean definition in the first sentence, headings that mirror how people actually ask ("what is...", "how do I..."), and facts an engine can quote without rewriting. Most people use AEO and GEO interchangeably, and honestly, so do we: they describe the same shift, just from slightly different angles.
Where a line gets drawn, it's this: AEO is about the page: is it shaped so an engine can extract a direct answer from it, while GEO is about the entity: does the wider web give the engine a reason to trust and name you at all. You need both. A perfectly-structured page from a business AI has never heard of still won't get cited; a well-known brand with vague, unquotable copy still won't get lifted. Found by AI measures both together, under one score, because a buyer asking ChatGPT doesn't care which acronym applies, see how in our ChatGPT-specific breakdown.
GEO vs AEO vs SEO: what's the difference?
SEO optimises to rank in a list of links; AEO optimises the page to be a direct, quotable answer; GEO optimises the entity to be trusted and named. They share foundations, a crawlable, fast, well-structured site helps all three, but they reward different things, and the shift between them is a shift in ambition.
Matching keywords
"waterproof rain jacket", you rank, the customer clicks, and decides for themselves among ten results.
Descriptive clarity
"lightweight, packable rain jacket with ventilation", your page is shaped so the engine can lift a clean, direct answer from it.
Justification & trust
"best-rated by Outdoor Magazine, 4.8 stars, 3-year warranty", the independent proof that gets you named, not just parsed.
Microsoft's own retail guide (2026) draws this exact ladder. Understanding gets you parsed; confidence gets you named. You need both, and the second one is earned off-page, which is why the comparison moments and your citation footprint carry the weight.
The shift: AI no longer returns a shortlist for the customer to evaluate, it evaluates for them and hands back a recommendation. GEO is how you become that recommendation.
What the research shows actually moves AI visibility
The foundational GEO study, Aggarwal et al., "GEO: Generative Engine Optimization," ACM SIGKDD 2024: tested nine content strategies across roughly 10,000 queries and nine datasets. Here's the lift each one measured, sorted from strongest to actively harmful:
"Including citations, quotations from relevant sources, and statistics can significantly boost source visibility, with an increase of over 40% across various queries."
— Aggarwal et al., GEO: Generative Engine Optimization, KDD 2024
This is why our six signals weight Citability and Brand so heavily: AI rewards content it can lift a clean, attributable fact from, published by an entity it already recognises. Notice what doesn't move the needle, length on its own does nothing, and the old SEO reflex of keyword stuffing actively costs you visibility.
What 2,354 real businesses prove
The Aggarwal study is a lab result. This is what we've measured live, across every business that's run a scan on this site, not a survey, not a sample, the actual scored corpus.
That readiness-vs-visibility gap is the whole framework in one number: the average business we measure has already done real work, a 67/100 foundation, and still gets named less than half the time an AI answers. Being ready and being recommended are not the same thing, and the gap between them is almost entirely earned off your own site.
Which signal actually explains that gap? We split our own corpus by score on each of the six signals and looked at the real visibility outcome. One signal dominates:
(average AI Visibility score, by Brand-signal bucket, across 3,268 scored scans)
Businesses scoring 70+ on Brand average 11x the visibility of those scoring under 40. E-E-A-T shows a real but smaller lift (48.1 vs 20.7, high vs low). Schema, on its own, shows almost none, 42–47.5 across every bucket, flat. That's not what we expected going in, and we're not hiding it: schema is table stakes, not a lever. It removes ambiguity for an AI that already trusts you; it doesn't make an AI trust you in the first place. Brand does that.
We're not the only ones finding this. Independent research on brand notability in generative search, Oruesagasti, "Algorithmic Trust and Compliance," 2026: found AI search systematically favours earned, third-party media over brand-owned content, the same pattern our own corpus shows. And on the measurement side, Schulte et al., "Don't Measure Once," 2026, make the case we build the product around: AI-search visibility has to be measured as a distribution across repeated queries, not a single snapshot, one lucky answer proves nothing.
In May 2026 Google published its own guidance on AI features in Search, and it is blunt: there are “no additional requirements to appear in AI Overviews or AI Mode”, and you “don’t need to create new machine readable files, AI text files, or markup” — explicitly including no special schema.org structured data. That is the same conclusion our corpus reached independently, before the guidance existed, and we had already published it above.
So take the honest version from us rather than the sales pitch from anyone else. An llms.txt will not get you into an AI answer. We still generate one because a handful of non-Google tools and agents read it and it costs you nothing, but we score it as a minor agent signal, not a visibility driver — and we say the same about schema. What actually moves the number is the thing Google’s own advice keeps pointing back to and our data ranks first: being a recognised entity that independent sources talk about.
One thing the guidance confirms rather than contradicts: Google describes a “query fan-out” technique — AI Overviews and AI Mode issue multiple related searches across subtopics before answering. That is exactly the layer we capture on every deep report, because it is where the decision actually gets made.
Common questions
Does AI actually recommend my business?
Possibly, and most owners have never checked. When a customer asks ChatGPT or Google's AI for the best business in your category and area, it names a few and omits the rest. A free scan shows whether you're named, who's named instead, and why.
How do I appear in ChatGPT and Google's AI?
Become a recognisable entity the AI can find, trust and quote: publish answer-shaped content backed by concrete facts, earn third-party citations (directories, reviews, press), add structured data, and make sure AI crawlers can read you. The six signals above are the checklist.
How is GEO different from SEO?
SEO earns a rank in a list of links; GEO earns a mention inside the AI's answer. Good SEO foundations help, but GEO additionally rewards entity recognition, citations and quotable content, and penalises keyword stuffing.
Is Answer Engine Optimisation (AEO) the same as GEO?
Effectively yes, most people use AEO and GEO interchangeably. Where a distinction gets drawn: AEO is about the page (is it structured so an engine can extract a direct answer), GEO is about the entity (does the web give the engine a reason to trust and name you). We measure both together, under one score.
What's the difference between SEO, AEO and GEO?
A ladder of ambition, per Microsoft's own retail guide: SEO matches keywords, AEO adds descriptive clarity so a page can be lifted as a direct answer, and GEO adds the independent trust signals that get you named rather than just parsed. See the full breakdown above.
What is AI search optimisation?
The plain-English umbrella term for GEO, AEO and every other name this category goes by: making sure AI systems, not just Google's blue links, find, trust and recommend your business. We use it, GEO and AEO interchangeably on this site for exactly that reason.
How long does GEO take to work?
For retrieval-based engines (ChatGPT search, Perplexity, AI Overviews) citation changes typically show within 4–8 weeks of publishing optimised content. Shifting what the base models "know" without web search takes longer, it depends on the next training cycle.
See your scores
Run the 60-second scan and find out whether AI recommends you, and exactly which of the six signals is holding you back.
LLM SEO, measured: what separates the businesses AI names from the ones it never mentions, from the same dataset this framework is built on.