How To Track Brand Mentions Across AI Models

Control the Session Conditions Personalisation quietly corrupts this. Run from a signed out session, or a fresh session with memory and history disabled, and do not use an account that has been researching your own company all week.

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.

Ask one final question before signing: what would you tell me if this is not working after six months? The answer reveals whether they have thought about failure, and an agency that has not thought about failure will not recognise it. llm seo

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.

What to Do First Run five prompts describing a purchase your best customer would be making, from a signed out session, and see what gets named and cited. Then check whether your product data survives with scripts disabled, and whether your name and identifiers are consistent across every listing you can find.

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 Price changes, new capabilities and discontinued services all silently invalidate published answers, and an outdated answer stated confidently is worse than no answer at all, because it can be quoted back at you by an assistant that has no way of knowing it is stale.

Why Real Questions Beat Generated Ones Questions produced by keyword tools are smoothed. They use category vocabulary, they avoid awkward specifics, and they tend to be the questions everyone has already answered.

One organisational habit makes this sustainable. Give the sales and support teams a single place to drop questions as they hear them, with no process attached beyond writing down the question in the customer's words. Anything more elaborate stops being used within a month, and a shared document with fifty verbatim questions in it is worth more than a formal intake process nobody completes.

This means a single answer is a sample. Being absent once is not evidence of a problem and being named once is not evidence of success, and treating either as a result is the most common analytical error in this field.

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 skill is knowing to sort cited domains by frequency, recognise which of them can be influenced, and understand that a competitor appearing in an answer is usually a story about a third party page rather than about their website. That is a different analytical habit from the one search built.

Why One Snapshot Proves Almost Nothing Generation involves randomness, and retrieval can return different pages between runs. The same prompt asked twice in a row can produce different companies in different orders.

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.

Check How They Handle Numbers Statistics circulate in this field faster than anyone checks them. A widely repeated claim about referral traffic growth turned out to rest on a sample of nineteen analytics properties. A frequently cited conversion comparison came from a vendor that sells the service.

Agree the Reporting Before You Sign Settle this in the contract rather than discovering it in month three. A useful monthly report contains the prompt set, the raw answers, which competitors were named, which sources were cited, what changed against last month, and what work was done that might explain the change.

Buy the technical audit if nobody on the team reads server logs, and buy the third party source work unless you already have a functioning public relations capability. Those are the two areas where the learning curve is steep and the cost of getting it wrong is highest.

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 discipline is in how you report their output. Every one of them samples: their own prompt set, their own infrastructure, their own run frequency. Their number is an estimate from a particular vantage point, not a count of what happened.