How Often Should You Re-Test Your AI Visibility

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.

Screenshots of favourable answers with no run count, which say nothing about how many attempts produced them. Impressions or traffic from unrelated channels included to fill a report. And activity described in the language of effort, such as ongoing optimisation, with no countable output attached.

The Objection, and the Answer to It Sales teams resist naming competitors and conceding anything, and the resistance is understandable. The counter is that the comparison is happening regardless, inside a model, using whichever sources it can find.

And read the raw text periodically rather than only the tallies. Changes in how you are described, from hedged to definite or from generic to specific, often precede changes in whether you appear at all, and no counting method will surface that. llm seo

Because there is no independent scoreboard in this channel, an engagement can run for a year on the strength of a number the supplier produces. That is an unusual amount of trust to extend, and it makes knowing what to check more important here than in any other marketing channel.

If you must change the prompt set, add new prompts as a separate cohort and keep the original series running unchanged. Editing the instrument retrospectively destroys the comparison you have been building.

Tracking this is genuinely awkward, and pretending otherwise is how most reporting in this field goes wrong. There is no console. Answers vary between runs. Referral attribution is inconsistent between assistants. Anyone handing you a single confident number has hidden a great deal of variance behind it.

If the only comparisons available are written by competitors and by review sites with incomplete information about you, that is the version being used. Publishing an honest one puts a source into circulation that at least contains your figures stated correctly, and honest treatment of where you lose makes the rest of the page more credible rather than less.

Where Analytics Can and Cannot Help Referral traffic from assistant domains does show up in analytics, and it is worth segmenting into its own report. Treat the numbers as a floor rather than a count, since some assistants strip referrer information and some traffic arrives looking direct.

Testing too rarely means you find out about a problem a quarter after it started. Testing too often means drowning in variance that looks like signal and reacting to noise. Both failures are common and the second is more expensive, because it produces work.

Be prepared for the internal objection that this sends people to competitors. Some of it will, and those are mostly people who would not have bought from you anyway. The trade is that the page becomes usable as an impartial source, which is worth considerably more than the small number of poorly matched prospects it redirects, and the sales team usually agrees once they see which enquiries stop arriving.

When to Test More Often Three situations justify a tighter loop. During an active campaign where you need to attribute a specific change, weekly runs on a subset of prompts are reasonable, provided you accept the variance.

So attribute it by name every time it appears in a report. A visibility figure presented without saying which tool produced it and how it was sampled will eventually be quoted back at you as fact by somebody who did not know it was an estimate, and that is a difficult correction to make in front of a board. llm seo

And in a fast moving category where competitors are actively publishing, monthly can miss a shift. Even then, keep the full set monthly and run a small subset more frequently rather than expanding everything.

What Honest Reporting Contains The prompt set, versioned and unchanged since last month. The raw answers, kept in full rather than summarised. Which competitors were named. Which sources were cited. What work was done. What moved, and the specific claim about which work caused it.

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.

One diagnostic shortcut is worth knowing. Ask the assistant to describe your company rather than to recommend one. If it produces an accurate description but will not recommend you, the record exists and the corroboration is thin, which points at third party sources. If it produces a vague or wrong description, the record itself is broken, which points at access and identity. Those two findings lead to completely different quarters of work, and the question that separates them takes ten seconds to ask.

Fragmented identity produces a specific symptom worth recognising: an assistant knows facts about you but attributes them vaguely, or confuses you with a similarly named business. The fix is dull consistency work across every place your name appears.