AI measurement
What should an AI SEO agency report before you renew its monthly package?
Good AI SEO agency reporting explains what changed, what was measured and what is still unknown. Check that evidence before the next renewal invoice.
AI SEO agency reporting should show the questions tested, captured answers and work actually completed. Keep mentions, citations, visits and qualified enquiries separate. Before renewing, compare those results with the original brief and the next proposed task. A rising score needs a visible calculation before it can support your buying decision.
A mention is only the beginning of the story
Your agency sends a chart showing more AI mentions. The sales team has not noticed more enquiries. Does that mean the work failed?
You cannot tell from the chart alone. A mention shows that a captured answer named your business. It does not show whether the reader was a suitable buyer, visited the website or contacted you.
Consider a possible journey: someone sees your company in an AI recommendation, searches for your brand later and then submits an enquiry. Your website analytics may record the search visit without recording the earlier recommendation. That gap matters when you discuss where the customer came from.
This is an illustrative journey, not an observed sequence from our clients. A short “How did you hear about us?” question can add useful context, but a person’s answer is still self-reported evidence.
Use two reporting pages
Keep captured answers and completed agency work on one page. Put visits, enquiries, qualification and sales evidence on the other. Connect them only where you have a defensible method; show the gaps where you do not.
Can you reproduce the question sample?
A useful comparison starts with the same core questions. Changing them every month can improve the percentage without improving the answers buyers actually see.
Keep the exact question, platform, date, language and relevant location context with each capture. Note sign-in or personalisation where it affects the method. If a platform does not expose the model or another detail, label it unavailable.
| Signal | Evidence | Limit |
|---|---|---|
| Mention | Captured answer names the business | May be irrelevant or inaccurate |
| Citation | Answer links to an identified URL | Does not establish a visit |
| Visit | Recorded session or request | Earlier influence may be missing |
| Enquiry | Contact reached the business | Still needs qualification |
| Sale | Recorded commercial outcome | Attribution needs its own method |
Keep new questions in a separate group until you have a comparable starting result. “What is SEO?” and “which agency can diagnose our WordPress indexing problem before a redesign?” concern different decisions. A mention for one cannot stand in for the other.
You do not need identical answers on every run. You do need a consistent method that explains repeated checks, missing answers and changes in the platform.
What can Search Console currently show?
Check which Search Console report your property actually has, then read what its numbers count.
Google’s generative AI performance report documentation, checked on 4 October 2026, describes impressions for supported Search AI features. It provides page, country, date and device views. An impression is a displayed link event, rather than a qualified enquiry.
Save the date range and filters with the export. Without them, next month’s comparison can be misleading. Some chart and table totals use different aggregation, so a mismatch should be investigated before it becomes a performance claim. If a report is unavailable in your property, record that fact instead of presenting a guessed total.
This is Google evidence. It does not measure all answers on ChatGPT, Perplexity or other independent services. Ask the agency to label each platform’s evidence separately.
How do you check whether the agency did the work?
Compare the report with the agreed deliverables and a dated change log.
A content task should identify the URL, what was changed and why. A technical correction should show the affected response or setting and its verification. A profile task should show the authorised destination and completed change.
Ask which recommendations remain waiting on your team. A transparent report may show unfinished work because a developer or subject expert has not approved it yet. Hiding that dependency makes next month’s plan less useful.
For a published guide, check the public page, links and author information yourself. Your agency should also be able to explain which assertions came from your team, primary documentation or an explicitly hypothetical example.
Our AI SEO agency selection guide helps put those acceptance checks in the brief before work starts.
What makes a score difficult to trust?
Start with the calculation. Without the number of questions and the actual sample, you cannot tell whether a higher score reflects improved answers or simply a different set of tests.
Suppose a hypothetical report says visibility rose from four mentions in twenty questions to six mentions in twenty questions. That is a change from 20% to 30%: ten percentage points, or a 50% relative increase. Both calculations describe the same sample. Keep those labels clear. “AI authority improved by 40%” means little without the formula, the underlying events and the reason the change matters.
Keep score changes separate from content accuracy. A business mentioned more often with the wrong location or service description has a correction problem even if the graph rises.
Ask how repeated runs, missing answers and changed platform behaviour are handled. The method should make uncertainty visible rather than erase inconvenient results.
What should your renewal decision consider?
Before renewing, check what was finished, whether you trust the evidence and what your buyers still need.
- Were the agreed tasks delivered and verified?
- Did the research cover the buyers and questions you care about?
- Can your team inspect the evidence and access its own accounts?
- Are enquiries and qualification measured honestly?
- Does the next plan address a clear remaining problem?
You might renew a useful programme before sales impact is conclusive, especially where measurement is incomplete. State that decision as an investment in defined work, not proof of a return the report cannot establish.
Alternatively, pause the recurring content batch and fund the technical or catalogue task blocking progress. A renewal meeting should leave you with a choice, rather than another chart and an automatic invoice.
Buyer questions
Does one ChatGPT citation prove our campaign worked?
It records one observed citation under the captured conditions. To assess a campaign, compare the agreed sample over time and review completed work, website evidence and commercial outcomes.
Should our brand be mentioned in every answer?
Relevance matters. A question outside your service or market does not become a useful target because your name can be inserted into it. Define the intended buyer questions.
Can an agency promise a permanent AI answer position?
Platforms change responses and retrieval behaviour. Ask for clear delivery commitments and measured observations instead of a permanent-position promise.
Sources and editorial method
Technical claims link to primary documentation in the relevant section. Buying worksheets are Digital Magma’s editorial recommendations. Examples are illustrative, not client results. This guide was prepared with AI assistance; no measured search volume or ranking outcome is asserted. Send corrections through our contact page.
Need a clearer brief for your next agency?
Bring the current report and the work you originally agreed. We can discuss what it establishes, what is missing and the next useful task.
Request an initial website review
Explore AI SEO services, compare SEO packages and read Google reviews.