Claude selects sources the way a careful researcher does under time pressure: it reaches for the few pages that answer the exact question, reads cleanly, come from somewhere credible, and agree with the rest of the record. That is the short answer, and most of what follows just unpacks why each of those conditions matters and how you satisfy them. If you want to be one of the names Claude returns when someone asks about your field, you have to look, to a machine, like the source a careful researcher would trust.
The mistake people make is treating this like old search engine optimization, stuffing keywords and chasing rankings. Claude does not hand the user ten links to sort through. When web search is on, it retrieves a set of pages, reads them, and synthesizes an answer that cites a handful of sources. You are not competing for position seven on a results page. You are competing to be one of three or four sources worth naming. That is a different, harder, more winnable game, and these seven signals are how you play it.
Retrieval first: you cannot be cited if you are not retrieved
Before Claude can weigh your page against anyone else’s, it has to find it. When web search is enabled, Claude pulls live results through a search layer, and if your page is absent from the indexes that layer draws on, nothing else you do matters. This is the first and most overlooked signal: retrievability. A brilliant page that no index surfaces for the query is invisible to the model, the same way a book that is not in the library cannot be checked out no matter how good it is.
That makes classic technical hygiene the price of admission. Your pages need to be crawlable, indexed, fast to load, and free of the barriers, aggressive bot blocking, content locked behind scripts, that keep retrieval systems from reading them. Plenty of businesses obsess over clever AEO tactics while their best content sits behind a setup that retrieval cannot parse. Fix the plumbing first. Being findable is signal one, and it gates the other six. A useful way to test it: search for the precise questions you want to own and see whether your pages appear anywhere in the results a retrieval system would pull from. If they are absent, you have found your first project, and no amount of polishing the content itself will matter until the pages can be reached. Retrievability is unglamorous and decisive, which is exactly why the businesses that skip it stay invisible while wondering why their good content never gets cited.
Signal two and three: direct answers and clean structure
Once retrieved, your page competes on how well it answers the specific question. Claude favors pages that address the query directly and early, not ones that bury the answer under a thousand words of preamble. If someone asks how a process works and your page opens with a crisp, accurate explanation of that process, you have given the model something it can quote with confidence. If the answer is scattered across the page or implied rather than stated, a competitor who said it plainly gets cited instead.
Structure is the close partner of directness. Clear headings that match real questions, short defined terms, a logical order from question to answer to detail, these make a page legible to a system reading fast. Think about how the page parses, not just how it reads. A well-built FAQ that answers the precise questions people ask, a definition stated in one clean sentence, a claim immediately backed by a specific number or example: these are the units a model lifts into an answer. The clearer the unit, the more quotable it is. Vague, meandering prose is hard to cite even when it is correct, because the model cannot extract a confident claim from it.
Signal four: authority the model can verify
The fourth signal is the credibility of the source itself. Claude leans toward domains and authors that carry markers of authority, established publications, recognized experts, sites with a track record in the subject. This is not snobbery; it is risk management. A model that cites a source is staking the answer’s reliability on that source, so it favors origins that are less likely to be wrong.
You build this signal slowly and you cannot fake it. A clear author identity with real credentials, an about page that states who you are and why you know the subject, consistent publishing in your niche, and recognition from other credible sites all add up to a domain a model treats as a safer bet. The businesses that get cited tend to be the ones that look, on inspection, like genuine authorities rather than anonymous content farms. Authority is earned in public over time, which is exactly why it works as a signal: it is expensive to counterfeit.
Signal five: corroboration across the record
Here is the signal most people miss, and it deserves its own name. Call it corroboration: the degree to which your claims are echoed by other sources the system already trusts. A model gains confidence in a piece of information when several independent, credible sources say the same thing. If your page makes a claim that three respected sites also make, the model can cite you with less risk. If your page is the only place on the internet asserting something, you are a thinner reed to lean on.
This is why third-party mentions matter so much for AI visibility, and why they are different from backlinks in the old sense. You are not chasing link equity for ranking. You are building a web of corroboration so that when a model encounters your claim, it finds the same claim, and your name, across the trusted record. Press coverage, expert citations, mentions in established publications, presence in the reference sources models lean on, these all thicken the corroboration around your entity. The Corroboration Web, the set of trusted places that confirm who you are and what you know, is the asset that turns an occasional citation into a default one.
Signal six and seven: freshness and entity clarity
The sixth signal is recency where it counts. For questions about current facts, prices, rankings, recent events, Claude favors sources that are up to date, because stale pages risk stale answers. Evergreen topics weight freshness less, but for anything time-sensitive, a recently updated, clearly dated page beats an authoritative but aging one. Keep your cornerstone pages current and stamp them with real dates.
The seventh signal is entity clarity: how cleanly the model can understand who you are and how you connect to the topic. Models reason over entities, the distinct people, companies, and concepts in their world, and you want to be a sharp, well-defined entity rather than a blur. Consistent naming across the web, a clear description of what you do, structured information that ties you to your subject, and unambiguous positioning all help the model file you correctly. The expert who clearly owns one specific subject is easier to retrieve and cite than the generalist who touches ten, because the engine can match a precise query to a precise entity with confidence.
A 30-day plan to act on the seven signals
Signals are useless without a sequence, so here is the order I would work them in over a month. Spend the first week on retrievability, because it gates everything. Confirm your key pages are crawlable and indexed, fix anything that loads slowly, and remove the scripts or blocks that hide content from machines. Run a real check: search for the exact questions you want to be cited for and see whether your pages surface at all. If they do not, nothing downstream matters yet, and this is where your time belongs.
Spend the second week on directness and structure. Take your three most important pages and rewrite the opening of each so it answers the core question plainly in the first screen. Add a clean FAQ that mirrors the precise questions people ask, with each answer stated in one or two self-contained sentences a model could quote without summarizing. Break long, meandering passages into defined claims backed by specific numbers or examples. You are not adding words; you are making the words you have easy to lift.
Use the third week on authority and entity clarity. Make sure every important page has a named author with real credentials, a clear about page that states who you are and why you know the subject, and consistent naming of your business and yourself across the web. The fourth week is corroboration: list the trusted sources that should mention you and do not yet, and start the slow work of earning those references through genuine expertise, contributed pieces, and relationships with people who write in your field. Freshness runs in the background the whole month, dating and updating your cornerstone pages as you go. None of this is a single sprint. It is a loop, and the businesses that win run it continuously rather than once.
Putting the seven signals to work
The seven signals stack, and they reinforce each other. Retrievability gets you in the room. Direct answers and clean structure make you quotable. Verifiable authority and a thick corroboration web make you safe to cite. Freshness and entity clarity make you the right, current, well-understood match for the query. None of these is a trick, and that is the point. Claude selects sources the way a thoughtful person does, so the durable strategy is to actually be the source a thoughtful person would choose, then make that fact legible to a machine.
Start by asking the questions your customers ask Claude, then read the answer and notice who gets cited. Those sources are your real competition, and the gap between them and you is a map. Usually the gap is not effort but precision: their page answered the exact question more directly, their claims were corroborated more widely, their entity was clearer. Close those gaps one signal at a time and you stop wondering why the model skips you. The work compounds, because every signal you strengthen makes the next citation more likely, and every citation thickens the record that earns the one after it.