AI discoverability
AI discoverability is whether an AI system can retrieve your page and then choose it as a source for its answer. Both halves have to hold: a page nothing can fetch is never a candidate, and a page that is fetched but repeats what the system already holds is not selected. This pillar sets out what is known about each half, as far as public evidence allows, and what a business can do about it.
What determines AI discoverability?
Five facts, each covered in depth in this pillar's articles:
- AI search is retrieval first. Google's generative features run on Search's core ranking systems; ChatGPT search depends on its own crawler reaching you and its rewritten queries matching you. If you are not retrievable, nothing downstream matters. Official recommendation
- One question becomes many. Systems issue multiple related searches for a single question, a mechanism called query fan-out, so pages compete on sub-questions they never targeted. Official recommendation
- Selection is volatile. The same question, asked twice, can cite different sources. Anyone promising a stable "AI ranking" describes a system that does not behave that way. Emerging
- Being cited and being used are different. A page can be listed as a source without shaping the answer, and shape the answer without prominence. Our study measures the two separately. Emerging
- There is no secret markup. No special schema, file or chunk size buys inclusion; see the folklore ledger. The platforms' own guidance rewards retrievability, relevance, and material that is not a commodity. Official recommendation
How is AI discoverability different from SEO?
Less than the name suggests, and in one place that matters. Everything under retrieval is the work ordinary search has always rewarded: pages a crawler can reach, an index that holds them, relevance to the question asked. None of that is replaced.
What changes is the last step. A results page offers ten links and lets the reader pick; an AI answer picks for them and names a handful of sources. The contest moves from a position in a list to a place in a written answer, and there is no position to hold. Emerging
So retrieval is the precondition and selection is the newer contest. Most of this pillar is about the second.
Our research in this pillar
We read the robots.txt of the 32 most-cited domains in our own measurement data. Naming AI crawlers has become normal; blocking them has not. The most-cited non-video source in the dataset blocks every automated client in four lines, and is cited anyway, which suggests access to the largest sources has moved into commercial agreements rather than crawl permission. Read the census.
The five facts above are claims, and claims on this site get tested rather than repeated. The Small-Business AI Citation Benchmark is the study that tests them, pre-registered before any data existed, and what we have measured so far is the running account of what the numbers currently support, including the readings that are still too small to report.
In this pillar
- How AI search finds sourceslive
- What makes a page citablelive
- Crawler controls: search vs traininglive
- How query fan-out changes content strategylive
- Appearing in ChatGPT searchlive
- Who is actually open to AI crawlers? (original research)live
- Google AI Overviewslive
- Google AI Modelive
- Perplexity: how it finds and cites sourceslive
- Gemini · Measuring AI citations · Browser agentsplanned
What bernard does about this pillar
Hosted sites get retrievability handled for them: static HTML, managed robots.txt with named AI search crawlers allowed, sitemaps, and verified logging of which AI crawlers fetch your pages. verified
No platform can guarantee citation, bernard included. Anyone who promises that belongs on the ledger. not controllable
Common questions
What is AI discoverability?
Whether an AI system can fetch your page and then pick it as a source when it answers a question. Both parts count: a page no crawler can reach never enters the pool, and a page that adds nothing the system already has does not get picked.
How is AI discoverability different from SEO?
The retrieval half is the same job: a crawler reaches the page, an index holds it, and it answers what was asked. Google's generative features run on Search's core ranking systems, so that work is not replaced. What is new is the last step: an AI answer picks a few sources itself, and not the same ones every time, so there is no rank to hold.
Is AI discoverability the same as AI findability?
The phrases are used interchangeably. This manual says discoverability throughout, and means both halves of the problem: being retrievable, and being worth selecting once retrieved.
Is there special markup that makes AI systems cite you?
No. No file, schema or chunk size buys inclusion, and the claims to the contrary are recorded in the folklore ledger with the evidence against them. What the platforms' own guidance rewards is being retrievable, being relevant, and holding something that is not a commodity.
Can any platform guarantee that an AI system will cite your site?
No, and that includes bernard. The same question asked twice can produce different sources, so there is no stable position for anyone to sell you. A platform can make you retrievable; selection is not in its gift.