AI Search Visibility

What changed

A growing share of the questions that used to produce a results page now produce a written answer. Someone asking “what should I do if I have been charged with a DUI in San Jose” may receive several paragraphs and one or two named suggestions, and never see a ranked list at all. Ten blue links degrade gracefully: position seven still exists. An answer does not: a firm is named or it is absent.

How an assistant assembles an answer

Broadly, three things happen, and each is a different intervention point.

Retrieval

The system fetches candidate documents, usually via a search index. If a firm’s pages are not crawlable, not indexed, or answer nothing anyone asks, there is nothing to retrieve. This is ordinary technical SEO doing new work, which is why the crawl and index stage matters more now instead of less.

Selection and synthesis

The model picks from what it retrieved and writes an answer. Sources that state things plainly, in self-contained passages, with clear attribution, survive this step better than sources that bury the answer in marketing prose. A page that answers the actual question in its first two sentences is easier to quote than one that opens with “For over 20 years, our dedicated team…”.

Citation

Some systems link their sources; some name entities without linking. Both are worth having, and the second is why entity consistency matters: the assistant has to be confident that “SBD Marketing” refers to one identifiable thing. See how AI assistants choose firms.

What actually helps

• Answer the question in the first paragraph. Assistants extract passages, not pages.
• Cover procedure and consequence queries, which is what people actually ask assistants. See criminal defense keywords.
• Be consistent about names, addresses and roles everywhere the firm appears.
• Emit structured data so the machine reading is unambiguous.
• Be crawlable to AI agents, check robots.txt does not block GPTBot, ClaudeBot, PerplexityBot, Google-Extended and the rest, unless that is a deliberate choice.
• Earn third-party mentions, which weigh more than self-description. See source bibliography.

The pages that earn citations

Some page types are cited by assistants far more often than others, and the pattern is consistent. Pages answering a procedural question, what happens at arraignment, how long a licence suspension lasts, whether a charge can be expunged, get cited because they answer something specific and can be quoted in a sentence.

Definition pages do similar work, since an assistant explaining a term reaches for a source that states it plainly. Pages giving a penalty range under a named statute are cited because the figure is checkable and attributable.

Pages that describe a firm’s approach, values or history are cited close to never. They contain nothing an assistant can attribute to anything but the firm itself, and they answer no question anybody asked. A firm wanting to be cited should count how many of its pages answer a question a stranger would type, and the answer is often very few.

Setting expectations about this channel

Assistant visibility is worth working on and it is easy to oversell, so the honest position needs stating. Nobody controls what an assistant says. There is no submission process, no ranking to buy, and no vendor with a relationship that changes an answer. What a firm can do is make itself easier to identify and harder to contradict, which raises the chance of being named without guaranteeing it.

The volume is also smaller today than the attention it receives. Most defense enquiries still arrive through ordinary search and the map pack, and a firm that neglects its profile to chase assistant citations has traded a working channel for a speculative one. The sensible order is to hold the pack and the organic results first, then treat assistant visibility as an additional surface the same work largely serves.

Any agency offering guaranteed placement in AI answers is selling something it cannot deliver. The defensible claim is that the underlying work, entity clarity, corroborated claims and pages that answer questions directly, improves the odds. That is the claim made here.

What people actually ask an assistant

The questions that reach an assistant differ from the ones typed into a search box, and the difference shapes what a firm should publish. Assistant questions are longer, more conversational and more likely to describe a situation than name a service. Somebody types “DUI lawyer near me” into a search box and asks an assistant what happens if they were arrested for a DUI last night and have court on Tuesday.

Situational questions of that kind are answered from pages that explain procedure, and the firms that get cited are the ones with a page describing what happens between arrest and arraignment in a named jurisdiction. A page that only lists services has nothing to contribute to that answer.

The second pattern is the shortlist question, where somebody asks for defense lawyers in a city. This is the harder one, since it asks the assistant to name businesses, and the default is to decline and point at a bar referral service. Being nameable there depends on entity clarity and independent corroboration more than on any page the firm writes.

Letting the crawlers in

AI systems reach a site through crawlers that are separate from the ordinary search crawler and separately blockable. GPTBot, ClaudeBot, PerplexityBot, CCBot and Google-Extended each control a distinct kind of access, and a site can be fully indexed by Google while being invisible to every one of them.

Many sites block these by default, sometimes through a security product or a hosting setting nobody chose deliberately. For a publisher protecting original work that can be a considered decision. For a law firm that wants to be named when somebody asks an assistant for a defense lawyer, it removes the firm from the channel it is trying to enter.

Checking this is quick: read the site’s robots.txt and look for the crawler names, then confirm the hosting layer is not blocking them independently of the file. Every site on this network publishes a robots.txt that admits these crawlers explicitly, which is the same position it recommends.

The surfaces are not one thing

Talking about AI search as a single channel obscures that the surfaces behave differently and draw on different sources. Google’s AI Overviews sit above the ordinary results and draw heavily on pages already ranking for the query, which makes conventional visibility the main route into them. A firm ranking nowhere for a term is unlikely to be cited in an overview of it.

Assistants that browse, including ChatGPT with search and Perplexity, run their own retrieval and cite what they retrieve. Their citation sets overlap with the ordinary results without matching them, and they favour pages that answer a question directly in a passage that can be lifted.

Assistants answering from training data alone are a third case and the hardest to influence, since nothing published today reaches a model already trained. What travels into the next training run is what independent sources say about a firm, which is the same corroboration that helps the browsing case. The practical implication is that the work is largely shared across all three, and the promises differ: overview citations follow from ranking, retrieval citations follow from answerable pages, and the training case follows from what others publish.

Measuring something this unstable

AI answers vary between runs of the same question, across accounts, and by the phrasing used, so a single screenshot of an assistant naming a firm establishes very little. Measuring it usefully means accepting that the output is a distribution instead of a value.

The method that survives scrutiny is to fix a set of questions, ask each one repeatedly on a schedule, and record how often the firm is named and what gets cited when it is. A firm named in three runs out of twenty has a real and small presence, and a claim of AI visibility built on one favourable screenshot is not a measurement. This network applies the same standard to itself: the figures published here are grid-measured search results with their scan dates attached, and no claim about assistant visibility is made in figures at all, because the honest version has not been measured to that standard.

What nobody can promise

See also: How AI assistants choose firms · SEO vs AEO · Criminal defense keywords