What Is AI Recommendation Share?

Someone types "best B2B payments platform" into ChatGPT. Stripe appears. You don't. Your Ahrefs Domain Rating is higher than theirs. Here's why that doesn't matter — and what actually determines who AI recommends.

What AI Recommendation Share is

AI Recommendation Share — or Match Share — is the percentage of buyer-intent questions in your category where an AI model recommends your brand as an answer, versus how often it recommends each named competitor.

It's a market-share metric, not a ranking metric. If ten founders ask ChatGPT "what's the best analytics tool for a small SaaS?" and PostHog is recommended in seven of those answers while Amplitude is recommended in three, PostHog's Match Share in that category is 70% and Amplitude's is 30%. Your own Match Share is whatever percentage of those answers name you.

The distinction that matters: being mentioned is not the same as being recommended. An LLM might list your brand as "one of several options" with a qualifier ("though some users find it complex"). That is a mention. Being named as the answer — with no hedging, in the first sentence — is a recommendation. Match Share measures the second, not the first.

Why SEO metrics don't predict it

SEO tools measure rank position on a search engine results page. AEO measures recommendation inside a synthesized AI answer. These are structurally different games, and a brand can be dominant in one and invisible in the other.

Traditional SEO rewards a page for outranking others when a user clicks a link. But LLMs don't return links — they return a paragraph. A model doesn't weigh your Domain Rating; it weighs whether your brand is a clean entity in its training data, whether the sources it trusts cite you for the specific buyer question, and whether the question itself has a canonical answer to which you're attached.

A brand with a DR of 82 can lose to a brand with a DR of 34 in ChatGPT if the smaller brand shows up in the review sites, discussion forums, and comparison pages the model synthesized its answer from.

This is why founders keep discovering the gap the hard way. Their SEO consultant tells them everything is fine. Their traffic looks stable. Then a customer tells them they asked ChatGPT which tool to use and the model named a competitor. That is not an SEO failure. It's an AEO failure — and it needs to be measured on its own terms.

The five signals AI models weigh

Across every scan we run, the same five signal categories determine whether a brand shows up as a recommendation. If your Match Share is low, the fix is almost always in one of these:

1. Recommendation presence. Does the model actually name your brand when asked the buyer-intent question? This is the output. Everything else is upstream of it.
2. Citation authority. Which third-party sources cite your brand for the category — and are those sources the ones the model draws from? A Product Hunt writeup, a Hacker News comment thread, a comparison post on a trusted blog, a Wikipedia entry. Not backlinks in the SEO sense — attribution in the LLM training sense.
3. Entity strength. Does the model understand your brand as a coherent, disambiguated entity? A company called "Match" competes for identity with the dating platform. A company called "Anchor" competes with Spotify's product. Entity strength is how cleanly a model can pull "your" knowledge apart from other same-named entities.
4. Commercial intent coverage. How many of the real buyer-intent questions in your category does your site actually answer, in language a model can extract? Not just "our product does X" — but "for teams of five to ten looking to do X, our approach is Y, and here's why." Coverage in the language of the buyer's question.
5. Technical readiness. Structured data, semantic HTML, canonical URLs, sitemap presence, robots directives that don't accidentally block AI crawlers. This is the price of admission — necessary but not sufficient.

The five stages of AI visibility

Every brand we scan sits somewhere on a five-stage progression. Knowing your stage matters more than knowing your raw score, because the highest-leverage move is different at each stage.

1. Invisible. The model has no meaningful representation of your brand as an entity. You will not appear in a recommendation regardless of the question. The fix is entity building — Wikipedia, structured data, canonical positioning content.
2. Detected. The model knows your brand exists but does not associate you with a specific buyer intent. You appear only when the question names you directly. The fix is commercial intent coverage — write the pages that map to the questions.
3. Understood. The model associates you with a category but recommends competitors more often. You get mentioned but not chosen. The fix is citation authority — get named in the sources the model already trusts for the category.
4. Trusted. The model recommends you as one of several answers. You appear with peers, without hedging. The fix is differentiation — help the model articulate the specific reason you're the answer for a specific slice of the market.
5. Recommended. The model names you as the primary answer when the buyer-intent question fits your slice. This is the target state for a defined product category.

Above Recommended sits Angel Mode — the state where a brand is recommended so consistently across models, categories, and phrasings that its Match Share compounds: mentions beget training data, training data begets more mentions.

How to measure your Match Share

The measurement problem is that "asking ChatGPT what it thinks of your brand" is not measurement. Models are non-deterministic. The same question, asked twice, can produce different answers. A single-shot check is anecdote.

A defensible measurement of Match Share requires four properties:

• Buyer-intent questions, not brand questions. "Best CRM for remote sales teams" is a buyer-intent question. "Is your CRM good?" is not. Measure the questions your buyers actually ask.
• Multiple models. ChatGPT, Claude, Gemini, and Perplexity have overlapping but distinct training and retrieval behavior. Averaging across them smooths model-specific quirks.
• Multiple runs per question. Non-determinism averages out over enough runs. Below five runs per model, you're measuring noise.
• Named competitor tracking. Match Share is a share metric. It requires counting not only whether you were named but which competitors were named alongside or instead of you.

What to do if your score is low

Every brand's fix depends on which of the five signals is the weakest — and which stage you're currently at. But three moves compound faster than the others regardless of category:

1. Write the questions. Publish a page for each real buyer-intent question in your category, with the question as the H1 and your answer directly beneath. Not a listicle — a real, opinionated answer. LLMs cite specific answers, not hedged ones.
2. Get named in the sources models draw from. This is unglamorous and irreplaceable. Comparison posts on category blogs. Reddit threads where you're referenced by name. Podcast transcripts. Product Hunt launches. It is not backlinks — it is being talked about in the corpus.
3. Disambiguate your entity. Add Organization schema. Claim Wikidata. Ensure your brand's Wikipedia (if one exists) is up to date. Make it trivial for a model to pull "you" apart from any other same-named entity.

The fastest way to know which of these applies to you is to measure. Everything else is guessing.