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.
On Third Party Tracking Tools Several tools now offer to monitor this at scale, and they save real time once your prompt set runs into the hundreds. They are worth buying for trend lines and for coverage you cannot manually sustain.
Making Any Model Safe Four clauses do most of the protective work regardless of structure. The prompt set and baseline archive belong to you and leave with you. Raw answers ship with every report. Scope is stated in countable units. And there is a defined review point with agreed criteria before the contract auto renews.
A retainer describing ongoing optimisation and strategic guidance with no countable deliverable is a subscription to a relationship. It may still be worth having, and you should know that is what you bought.
Descriptions of your business moving from hedged to definite, which you can read yourself in the raw answers. Third party sources that previously described you wrongly now describing you correctly. And an increase in the fraction of runs naming you on buying intent prompts specifically, reported with run counts visible.
If the budget is substantial, add the earned coverage work, which is the slowest and most expensive component and the one you genuinely cannot do quickly on your own. Buying that first, before the cheap fixes are done, is the most common way money gets wasted in this field. brand mentions in ai answers
Ask what was done, not what happened. If listings were corrected, pages rewritten and outreach attempted, and the numbers are still flat, that is information about the market. If none of it happened, the numbers were never going to move.
One structural decision saves a lot of trouble later. Keep the raw answers in plain text files named by date, assistant and run number, rather than pasting them into a document that gets reformatted. Six months in you will want to search across every run for the first appearance of a competitor or a source, and a folder of plain files supports that while a slide deck does not.
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.
Also decide up front who owns this. Measurement that belongs to everyone gets run inconsistently, the conditions drift, and the series becomes uncomparable within two quarters. One named person running a modest set reliably produces more usable information than a sophisticated programme with no owner.
The Signals That Mean Something Four things are hard to fake and worth watching closely. Your own pages beginning to appear in cited sources, which is directly observable in any assistant that shows citations.
Reading Retrieval Rather Than Rankings Search reporting trained everyone to read a position number. This channel produces a body of text and a list of sources, and the useful information is mostly in the sources.
What a Local Business Should Do This Month Run five prompts asking for a business like yours in your town, from a signed out session, and record who gets named and what gets cited. Then fix every listing on the sources that appeared, starting with the phone number and address.
Record the conditions alongside the results: which assistant, which model version if visible, whether web access was on, the date and the run number. When a result changes sharply, the conditions log is usually what tells you whether the world changed or your setup did.
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.
An agency doing the work sends these the same day, because they already exist as a by-product of the measurement. One that does not will explain that the platform does not export in that format, or that the data is summarised in the dashboard.
Write it once, covering the category question, the problem question, the comparison question, the competitor question and the branded question. Fifty is a workable minimum. Then freeze it, and if you must add prompts later, add them as a separate cohort so the original series stays comparable.
There is also a straightforward test that costs nothing and tends to end the debate internally. Ask an assistant the question your best customer would have asked before they found you, and read the answer out in the next management meeting. brand mentions in ai answers
When to Change Supplier Three conditions justify it individually. Raw answers cannot be produced on request. The prompt set has been changed without disclosure, which invalidates every comparison in every report you have received. Or two quarters have passed with the agreed inputs completed and no movement on citation presence, accuracy or source coverage.