Ask sales to note the question asked on every call for a month, in the prospect's words rather than paraphrased. Export support tickets and sort by frequency. Pull the query report from Search Console. And read the first message from inbound enquiries before anyone has reshaped it.
There is a defensible way to measure this. It produces less certainty than a paid media report and considerably more than a visibility score, and it has the advantage of surviving scrutiny. ai search optimization
Expect the vocabulary to keep shifting, and expect new terms to arrive with each wave of positioning. The underlying work has been stable since these systems started retrieving live sources, and it is the work rather than the name that you are buying. ai search optimization
Establish What They Will Not Promise Nobody controls what a model says. There is no submission process, no ranking factor to buy, and no relationship with a provider that reserves you a place in an answer. Any guarantee of a specific position or mention is describing something the agency cannot deliver.
What to Do About llms.txt and Similar Files Proposals for machine readable files aimed specifically at language model consumers appear periodically. Adoption is inconsistent and support varies by provider, so treat these as low cost and speculative rather than as a requirement.
Blocking these is therefore not one decision. Turning away a training crawler is a defensible editorial position. Turning away the agent that fetches pages at answer time removes you from answers entirely, and the two are frequently confused.
Accuracy Beats Coverage The most common real defect is not missing markup, it is markup that disagrees with the page or with the rest of the web. A founding year in your schema that differs from your about page. A logo URL that returns a 404. A contact point nobody monitors.
The Types That Rarely Earn Their Keep Elaborate breadcrumb hierarchies, speakable markup, deeply nested item lists and most of the specialised types outside their intended vertical produce little observable difference in how a brand is understood or recommended.
Ask What They Cannot Measure A competent practitioner will volunteer limitations before you ask. Assistant answers vary between sessions. Referral attribution is inconsistent. Some assistants cannot be measured reliably at all. Sample sizes in the published research are small.
In practice it is used to mean roughly the same thing as generative engine optimization, occasionally with a stronger emphasis on training data and brand presence in the underlying corpus rather than on live retrieval.
The same applies to limitations. Stating plainly what you do not do, what size of job you decline and which situations suit a competitor produces the constraint statements that models lift as impartial facts.
The defensible position is to spend an hour on it if you like, and to spend the rest of the week on the things every system already reads: accessible pages, accurate Organization markup, consistent identity and content a machine can quote.
Treat markup as something with a maintenance cost rather than a one off implementation. Prices change, people leave, products are discontinued, and structured data quietly keeps asserting the old version long after the visible page has been updated. Adding a schema review to whatever process already updates your pages costs minutes and prevents the most damaging failure mode, which is confidently stating something that is no longer true.
Revisit the answers when the business changes rather than on a content schedule. Price changes, new capabilities and discontinued services all silently invalidate published answers, and an outdated answer stated confidently is worse than no answer at all, because it can be quoted back at you by an assistant that has no way of knowing it is stale.
Generative Engine Optimization The broadest of the three in common use. It refers to being visible in systems that generate an answer rather than returning a list, which covers assistants, AI summaries on results pages and any interface that synthesises rather than links.
Real questions are messy, specific and frequently uncomfortable. They ask about price, about limitations, about whether you can handle a particular awkward situation. That specificity is exactly what makes an answer quotable, because it matches the shape of a real query rather than a generic one.
The Blocks Nobody Chose Most blocking discovered during audits was never a decision. A disallow copied from a template. A staging rule that survived a migration. A security plugin with an aggressive default. A content delivery network setting labelled bot protection with a switch nobody has looked at since launch.
Vague answers about digital PR are a warning sign. Good answers are concrete: they have read your baseline source list, they know which platforms allow corrections, they have a view on which comparison articles are worth approaching, and they will tell you which ones are out of reach.