How to Measure AI SEO Performance
Build a report that separates search data, AI citations, visits and buyer actions, with clear limits and next steps.
Use four kinds of evidence to measure AI SEO work. Record search results, the AI citations you really observed, site visits and real buyer actions. Then add the cost of the work. Label where each number came from and what it can show. A mention is not a sale, and missing data is not proof of hidden value.
Start with a business question you can act on. You might want to know whether the pages you changed brought more good-fit leads. Name the pages, the owner and the dates you will compare. Decide what a good-fit lead means before you count it. Keep the work you completed apart from proof that results improved.
Which search metrics should you log?
Use Search Console to log clicks, impressions, CTR and average position. CTR is clicks divided by impressions. Average position is the mean for your topmost result in the view you chose. Keep the dates, search type and filters with each figure. A site-wide mean is not the rank of all keywords.
Search results. Start with the same search type and matched dates. Log clicks, impressions and CTR. Keep the query or page view and filters with the export. Then you can repeat a check with the same scope.
Position. Look at the query and page you care about. The site-wide mean does not tell you that page's rank for all searches. Tables can group the figures in ways that change the mean. Check the group that fits the buyer question.
Query rows. Count the rows you can see and note the scope. Check which pages appear for them. The list shows queries in this view, not all keywords you could target. Ask if new queries fit the business. A new mix can change the site-wide mean.
Visits and actions. Check GA4 sessions and the key events you choose to log. Compare them with real leads and sales. A search click and a session are not the same unit. A form fill is not always a good-fit lead. Keep revenue and costs apart so you can work out profit.
How can you track AI citations?
Use each source for what it can show. Search data, counts of cited pages and a small question log are not the same thing. Keep their scope clear. None shows all AI answers or the full path to a sale. Do not add unlike counts and call that a market share.
Google AI impressions. Use Search Console's Generative AI performance report for Search. It covers AI Overviews and AI Mode. It shows link impressions by page, country, device and date. These are not clicks or sales. Search Labs tests are not included. The report may be absent if there are too few impressions or the site is excluded. Overall Web totals still combine kinds of search results.
Recorded visits. GA4's session source and medium show the traffic-source details it logged. (direct) / (none) means it has no clear referral source. More than one thing can cause that. Keep the source gap clear. Do not count all of it as AI.
Cited pages. Bing's AI Performance report counts source citations for Copilot, Bing and select partners. Use it if your site has access. It does not give rank, placement, clicks or sales. Its grounding queries are a sample. Keep that scope with the count.
Your sample. Keep a log of the questions you test. Save the service, date, place, device, login state and settings with each answer and its source links. Log brand names apart from page links. Say how many checked answers cite your page out of all answers checked. That rate is for your sample, not all users.
Keep what the data shows apart from your view of why it changed.
What should your SEO report show?
Our suggested report has three sections: scope and data quality, work and observed results, then the next decision. This keeps the evidence close to the action it can support. A list of pages published is a work record. It is not proof of more leads. For each finding, show what changed, how you checked it and what you still need to know.
Scope and quality. State the period, filters, source and sample size. Name the owner and the data export or saved view. Mark missing values as missing. New Search Console figures may still change. Keep an unavailable value apart from a measured zero.
Work and results. List the exact pages and versions changed and which checks passed. Keep search, citation, visit, lead, sale and cost figures in their own units. Allow for the time between a visit and a sale. Count shared costs once and avoid counting the same sale twice.
Next decision. Separate a possible cause from a cause supported by evidence. Name the next repair or check, its owner and a review date. Say what would justify keeping, changing or stopping the test. If the cause is unknown, explain which check could reduce that gap.
For related agency reporting questions, read is your SEO agency actually working. For wider business-case context, see is SEO still worth it in 2026. These related guides do not establish the cause of a change in your own data.
How do you check why the numbers changed?
A rise in impressions with flat clicks needs a check, not an automatic AI explanation. Compare the same query or page over suitable periods. Look at position, query mix, country, device and changes in demand. Check known site edits and whether visits and buyer actions were logged as expected. A rise in brand searches alone does not prove AI caused it. Falling traffic matters too, but a lack of brand growth does not prove there was no AI exposure. Name the next check that could help explain the change.
How often should you review AI SEO metrics?
Choose a review rhythm that fits the data and the work. A clear site or tracking fault needs prompt attention. Trends need enough data for a useful comparison, and sales may take longer than a visit to appear in the records. Name an owner and a review date. Recheck after relevant site or tracking changes. Do not treat each short-term move as a trend, or wait for a monthly report to inspect a known fault.
Which tools can you use?
Start with Search Console and GA4. For Google AI link impressions, use the dedicated report for Search. If it is missing, check its help and the site's Search generative AI control under Settings. The default is inclusion, or the parent choice for a child property. Do not treat a missing report as a measured zero. Use Bing's AI report if your site has it. Keep the exports and a small question log. Check the date, source and counting rules before you use a total. For any paid tool, ask which search services, places and questions it checks. Find out what you can export. Test a small known sample before you rely on its totals.
Sources
Not sure your SEO reporting captures the full picture in 2026? Get a free Brand & Tech Assessment. We will show you what you might be missing.
Book a free Brand and Tech Assessment to see exactly how we would grow your organic visibility.
The Through The Glass Creatives Difference
There is a reason brands choose Through The Glass Creatives for work like this. The team is led by Ravve Jay Prevendido. He is the creative director behind OWWA, Nuvia, and 100+ brands. Alongside him is Mherie Vic Palomo-Prevendido, a growth and brand strategist. Together, TTGC builds a managed system that compounds over time, not a one-off project or a ticket queue. When the outcome truly matters, Mherie, Ravve, and the TTGC team are the ones to trust with it. Book your free Brand and Growth Assessment.
