Evidence gaps: what makes an answer pass you over

Key takeaways

• An answer can pass you over in two ways: it leaves you out, or it names you without putting you forward. The fixes differ.
• We sort the reasons into twelve evidence gap classes. Each one points at a specific kind of work, and one lost answer can carry several.
• Treat the classes as a guide to where to look, not a verdict. Read the answer itself before you commit effort.

When an AI answer puts a rival forward and not you, the natural reaction is to ask what the rival did. A more useful question is narrower: what was missing from the evidence about you, in this answer, for this question? That question has answers, and each answer is a piece of work someone can own.

We sort the reasons an answer passes a brand over into twelve evidence gap classes. Eight can be read from what an answer contains: who it named, in what role, which sources it cited and what claims it made. Four depend on information an answer does not contain, such as whether your pages can be crawled at all, and need a separate check.

A caution before we start. These classes are working rules written to make lost answers easier to act on. They are a strong prompt for where to look, not a final diagnosis, and we use them alongside reading the answer itself, never instead of it.

Two ways to lose an answer

Every gap starts with a lost answer, and there are two shapes of loss.

Absent. Your brand is not named in the answer at all. The question was asked, other vendors were put forward, and you were not part of the conversation.

Named, not the primary recommendation. Your brand appears, but someone else is put forward as the fit. You might be an alternative, part of a neutral list, compared or mentioned in passing.

The two shapes call for different work. Being absent is usually a problem of connection: nothing ties you to the question. Being named but not chosen is usually a problem of support: you are connected, but the evidence for you is thinner, vaguer or less independent than the evidence for the brand that was chosen. Several of the classes below apply to one shape only.

Class	What it looks like	First fix
Entity	Your site is cited, but your brand is not named	Make the brand name explicit on the page and in its title
Coverage	You are absent, and nothing on your domain was used	Publish a page that answers the question in the buyer's words
Specificity	Named, not chosen; the chosen brand has a specific supporting claim and you have none	State the specific fit the buyer asked about
Proof	Named, not chosen; the chosen brand has a supporting source and you have none	Give the claim a citable page
Corroboration	Every source supporting you is your own	Earn one independent source for the claim
Consensus	The answer is mixed about you	Address the critical point directly
Reputation	The answer discourages you or carries a critical claim	Fix or correct the issue behind it, and say so
Authority (breadth of support)	The chosen brand is backed by more distinct sites than you	Broaden independent support across more sites
Eligibility	The question asks for something you may not offer	Check the fit honestly before writing anything
Retrieval	Your evidence may not be reachable at all	Check crawling, indexing and rendering
Format	Your evidence exists but is hard to lift a claim from	Restructure the page around plain, quotable statements
Freshness	Your evidence may look out of date	Date your pages and keep the important ones current

Gaps of connection: entity and coverage

These two explain most answers where you are absent.

Entity. Your own site is cited as a source, but the answer does not name your brand. It sounds paradoxical. An answer uses your integration guide to explain how warehouse connectors work, then names two rivals as the vendors to consider. Usually the page explains the topic well but does not make clear, in the text an answer would draw on, who is offering what. The fix is to make the brand explicit: the name in the page title and opening, a plain sentence that says your product does the thing, and consistent use of one name rather than a product name on one page and a company name on the next.

Coverage. You are absent, and nothing on your domain was used to write the answer. The answer was built entirely on other people's pages. This is the plainest gap: as far as the answer is concerned, you have nothing to say about this question. The fix is a page that answers the question directly, in the words a buyer uses, with the specific facts the answer would need. Not a product page with the question somewhere on it, but a page that is about the question.

Gaps of support: specificity, proof, corroboration and authority

These four explain most answers where you are named but someone else is chosen. They are about the quality of the evidence behind you, compared with the evidence behind the brand that won.

Specificity. The chosen brand carries a specific supporting claim in the answer and you carry none. "Pennant Metrics documents UK data residency and audit logging." Against that, "your brand is also an option" will lose. The fix is to state, in public, the specific fit the buyer asked about. If the question names regulated reporting, say exactly what you do for regulated reporting, with the details a compliance lead would want.

Proof. The chosen brand has a source that supports it in the answer and you have none. The answer could point to something for the rival and nothing for you. The fix is to give each important claim a page that can be cited: a security overview, a documented integration, a certification with its scope, a case study with named and checkable facts. A claim that exists only in a sales deck or behind a form cannot be cited.

Corroboration. You have supporting sources, but every one of them is on your own domain. Nobody else is saying it. An answer weighing a self-description against a rival's claim repeated by a partner directory and an analyst note has an easy choice. The fix is to earn one source beyond your own domain for each claim that matters: a complete partner or marketplace listing, a directory entry, an accurate mention in a comparison. You cannot write these, but you can supply the accurate material that makes them likely.

Authority (breadth of support). The chosen brand is backed by sources on more distinct sites than your independent support, and by at least two. It counts breadth: how many different sites back the rival compared with how many back you. It does not judge whether any site is more authoritative than another. The fix is to widen independent support across more sites rather than piling more material onto one.

Gaps of description: consensus and reputation

Sometimes the evidence about you exists and is found, but it does not say what you would like.

Consensus. The answer is mixed about you: its tone is mixed, or it carries both supporting and critical claims. "A strong reporting tool, though some teams find setup slow." The fix is not to drown out the critical point but to address it. If setup was slow and is now faster, publish what changed, with specifics, and make sure partner and review pages can find the current position. If the criticism is fair, decide whether to fix the product or to be clear about who you suit.

Reputation. The answer discourages you, is negative about you or carries a critical claim about you. This is the most serious class, because a discouraged mention can do more harm than no mention. The fix starts with the substance. Find the sources behind the claim. If they describe a real problem, the work is the product or the service, then a clear public account of what changed. If they are out of date or wrong, publish the correction where the claim lives and ask the owners of the sources to update them.

Gaps that need a separate check: eligibility, retrieval, format and freshness

Four classes cannot be read from an answer, because the answer does not contain what they need. Each needs a check of something outside the answer.

Eligibility. The question asks for something you may not offer. If buyers ask for on-premises deployment and you are cloud only, no amount of content will make you a fair recommendation, and you should not try. The check is an honest inventory of what you actually do. The fix is either a product decision or a decision to concentrate on the questions you do fit.

Retrieval. Your evidence may not be reachable. A page blocked from crawlers, missing from search indexes, hidden behind a login or only rendered by scripts a crawler does not run cannot be found, read or cited. The check is technical: crawl access, index status and whether the important text is in the page a crawler receives. The fix is usually quick once found.

Format. Your evidence exists but is hard to lift a claim from. The key fact is buried in the middle of a long paragraph, held in an image, split across tabs or written in language so general that no specific claim can be taken from it. The check is to read the page as an answer would: can you find the one sentence that answers the buyer's question? The fix is plain structure: clear headings that match the questions, short statements of fact, tables for comparisons.

Freshness. Your evidence may look out of date. An undated page, or one dated three years ago, about a capability that has moved on gives an answer little reason to trust it over a rival's current one. The check is whether your important pages carry reliable dates. The fix is to date them, review them on a schedule and update them when the facts change.

How to work through your gaps

Knowing the classes is useful. Working through them in a sensible order is what changes answers.

1. Start from lost answers on questions that matter. Take your highest weighted buyer questions and list the answers where you were absent or named but not chosen.
2. Read each answer before classifying it. Note who was chosen, for what reason and on what sources.
3. Assign the classes that apply. Several will often apply to one answer. Where none of them fits, record it as unclassified rather than forcing it into the nearest class.
4. Count across answers. A class that appears across many families and many questions is a pattern worth a programme. A class that appears once is worth a look.
5. Turn each pattern into owned work. One gap, one piece of work, one owner, one date, and the questions it should move.
6. Ask the same questions again once the work has shipped and had time to be found, and compare like with like.