What Is AI Search Visibility?

What changed

A growing share of questions that used to produce a results page now produce a written answer. Someone asking “what should I do if I've been charged with a DUI in San Jose” may get several paragraphs and one or two named suggestions, and never see a ranked list.

Ten links degrade gracefully: position seven still exists and still gets clicks. An answer does not. This is the structural difference and it is why the surface is worth separate attention.

How an answer gets assembled

Retrieval. The system fetches candidate documents, usually through a search index. Pages that are not crawlable, not indexed, or answer nothing anyone asks cannot be retrieved. Ordinary technical SEO doing new work.

Selection and synthesis. The model picks from what it retrieved and writes. Sources that state things plainly, in self-contained passages, survive this better than sources that bury the answer in marketing prose.

Citation. Some systems link sources; some name entities without linking. Both are worth having, and the second is why entity consistency matters.

What actually helps

• Answer the question in the first paragraph. Assistants extract passages, not pages.
• Cover procedure and consequence questions: what people actually ask assistants.
• Be consistent about names, addresses and roles everywhere the firm appears.
• Emit structured data matching the visible page.
• Check robots.txt does not block AI crawlers — GPTBot, ClaudeBot, PerplexityBot, Google-Extended: unless that is a deliberate choice.
• Earn third-party mentions, which weigh more than self-description.

Why independence matters here

Retrieval systems weight corroboration across sources that do not share a publisher. A claim appearing only on the claimant's own site is weak evidence. This has a direct implication for this publisher: SBD Marketing runs ten domains on this topic, and under this frame they are one source, not ten. That is stated on every page of every one of them. See the network page.

What changed

An increasing share of defense research begins with an assistant rather than a results page, and the shift is concentrated at the research end. Questions about procedure, penalties and what happens next go to an assistant; the urgent call at eleven at night still goes through the map pack.

The difference that matters is what the reader sees. Ordinary search shows several firms and lets the searcher choose. An assembled answer has already chosen, and a firm left out of it is absent altogether.

The surfaces are not one thing

Three different things get called AI search and they behave differently.

AI overviews sit above the ordinary results and draw heavily on pages already ranking for the query, which makes conventional visibility the main route in. A firm ranking nowhere for a term is unlikely to be cited in an overview of it.

Assistants that browse run their own retrieval and cite what they find. They favour pages answering a question directly in a passage that can be lifted whole.

Assistants answering from training data are the hardest case, since nothing published today reaches a model already trained. What travels into a later training run is what independent sources say about a firm.

What actually helps

Four things, and none of them is a placement anybody sells.

Crawler access. GPTBot, ClaudeBot, PerplexityBot and Google-Extended are separately blockable, and a site fully indexed by Google can be invisible to all of them through a robots.txt line or a hosting setting nobody chose deliberately.

Entity consistency. One firm name, one address, one phone number, everywhere. An assistant finding contradictory information about a firm resolves it by saying less, and a vaguer answer usually names somebody else. See entity and NAP.

Answerable passages. A question as a heading, answered in the sentence beneath it, with a specific figure and a named statute where one applies. A sentence saying penalties vary by circumstance is true and no assistant will cite it.

Independent corroboration. A claim only the firm makes gives an assistant nothing to stand on, and the more emphatic the claim the more conspicuous the absence.

What nobody can promise

Nobody controls what an assistant says. There is no submission process, no ranking to buy, and no vendor whose relationship changes an answer. An agency guaranteeing placement in AI answers is guaranteeing something outside its control.

Measurement is different in kind too. Answers vary between runs, across accounts and by phrasing, so a single screenshot establishes very little. The honest method is to fix a set of questions, ask them repeatedly on a schedule, and record how often the firm is named. That produces a distribution rather than a position.

This dictionary publishes no figures about assistant visibility, because that work has not been done to the standard applied to the search figures. A screenshot would not meet the standard this network asks of others.

Answer check

Apply the answer to What Is AI Search Visibility? by naming the actual firm, market and decision date. Identify the source or asset that can settle the question. Keep the observation in its native unit and write down any missing field before comparing options.

The short answer supplies a working rule. It does not guarantee a ranking or business result. Recheck the rule when the market, source or measurement method changes, and retain the earlier decision when the change is material. Name the person responsible for that review.

See also
Answer engine optimization · SEO vs AEO · Entity