How to Measure SEO Success When Search Uses AI

Start with the result your business needs

A key performance indicator, or KPI, is a measure tied to a goal. For a service business, the goal might be qualified enquiries and profitable bookings. For a retailer, it might be completed orders and the profit they generate.

Choose a small set of measures that explain progress towards that goal. If your team cannot say how a number will affect a decision, it probably does not belong in the main summary.

These stages are a reporting framework, not a fully traceable customer funnel. One person can see several links, visit repeatedly and buy later through another route.

What Google Search Console measures

The Search results Performance report shows impressions, clicks, click-through rate and average position. Use relevant page, query and market views to investigate performance, rather than judging the whole business by one average.

Google’s Generative AI performance reports launched on 3 June 2026, with worldwide rollout completed by 31 August. The Search report provides a dedicated view of impressions in AI features, including AI Overviews and AI Mode. It shows pages, countries, devices and dates.

This AI visibility data is already included in overall Search reporting. Do not add it to the overall total as an extra audience. The documented AI report does not provide an Overview-only breakdown or separate AI clicks, click-through rate and revenue.

Google’s Generative AI reporting announcement

Search impressions are not unique people. The report measures displayed links, rather than every unlinked mention of your brand. Its help documentation also notes that a report may be absent when impressions are insufficient or the site is excluded from generative AI features. Search Labs experiments are not included.

Google’s help for the Generative AI Search report

What Bing AI Performance measures

Microsoft launched AI Performance in Bing Webmaster Tools as a public preview in February 2026. It reports citations across Microsoft Copilot, AI-generated Bing summaries and selected partner experiences; it does not cover every AI system.

Total Citations counts displayed source citations. Average Cited Pages measures the average number of distinct cited pages per day. Page-level counts identify referenced URLs, while grounding queries provide a sample of the phrases used to retrieve supporting content.

These figures are not visits, rankings or sales. A grounding query is not necessarily the customer’s exact original prompt. Keep the platform coverage and preview status visible in your report.

Microsoft’s introduction to AI Performance

Use the newer insights for context

Microsoft’s June 2026 preview update introduced Intents, Topics, Citation Share and Compare. Intents classify query context; Topics group related grounding queries; Compare helps examine changes between periods.

Citation Share is your site’s percentage of citations for a particular grounding query across the cited sites. It is not a share of traffic or sales, and does not reveal competitor domains or assign a quality score. Intent and topic labels can be broad during the preview.

Use these views to identify questions worth investigating. An informational topic might justify improving an explanation; it does not prove that the audience wants to book your service.

Microsoft’s newer AI visibility insights

What GA4 AI Assistant visits measure

Google Analytics introduced the AI Assistant channel in May 2026. Recognised assistant referrers are classified under that channel, with an ai-assistant medium and an (ai-assistant) campaign value. This helps identify recorded visits from supported assistants.

Crucially, GA4’s default channel definitions exclude Google AI Overviews and AI Mode from AI Assistant. Visits from those Google Search features belong in Organic Search. Gemini assistant referrals and Google AI search visits are therefore not the same category.

Google Analytics product updates

GA4 default channel definitions

For monthly visits, use sessions with Session default channel group in Traffic acquisition. First user default channel group answers a different question about initial acquisition. Keep the scope consistent when comparing months.

A session is a visit, not a unique customer. GA4 can record behaviour only through the measurement available to it. Missing referrer information, consent choices or implementation problems can leave some journeys unobserved or classified differently. The AI Assistant row is not a complete count of everyone influenced by AI.

GA4 Traffic acquisition report guidance

Measure enquiries and sales separately

Measure completed enquiries or purchases. A telephone-link click, opened form or started checkout shows intent, but does not establish that a conversation or sale happened.

GA4 calls important measured actions key events. Test that enquiries are recorded after successful submission without duplicates. If you sell online, test purchase tracking, currency and value.

Use your customer relationship management system, booking records or sales ledger to identify genuine enquiries, suitable leads and paid work. Exclude spam and duplicates consistently.

State your attribution rule. Sales associated with an enquiry’s recorded Organic Search source are not necessarily sales caused entirely by SEO. Ordinary attribution reporting cannot establish that causal claim.

A worked monthly report

The following figures are fictional. They show how a small garden maintenance business could compare August and September 2026 without combining incompatible measures.

The business uses the same website properties, filters and event definitions in both months. GSC figures use its Web search reporting scope. GA4 rows use session channel groups. Currency is GBP. August has 31 days and September has 30, so the owner also checks daily trends before interpreting the changes.

Exposure and visits

September’s overall GSC impressions remain 120,000. Adding 18,000 to report 138,000 would count overlapping reporting twice. Adding Bing citations would introduce another different unit, not produce a combined audience total.

Google clicks and GA4 sessions need not match: they have different definitions and collection methods. Keep both rows labelled. Do not divide the 120 assistant sessions by the 480 Bing citations to invent a click-through rate; their platform coverage differs.

Enquiries and commercial outcomes

In this example, recorded completed-enquiry events have been reconciled with unique enquiries. Qualified enquiries are confirmed by the team. Sales and revenue refer to work completed in each month, including work from earlier enquiries. Sales use the last recorded channel of the originating enquiry in the CRM, not a separate GA4 revenue attribution model.

Six enquiries from 120 assistant sessions gives an illustrative enquiry-per-session ratio of 5%, compared with 52 from 2,600 organic sessions, or 2%. The counts are small, so the owner checks quality and longer-term trends before changing the budget. These ratios are not a measure of AI visibility converting directly into sales.

Do not calculate a same-month lead-to-sale rate from these sales rows: some completed jobs came from earlier enquiries. To measure that rate, group leads by the month they arrived, allow a consistent follow-up period and count the wins from that same group.

What the owner can reasonably conclude

Visibility increased in the recorded Google and Bing measures. Organic Search visits and qualified enquiries also grew. Recognised assistant visits produced four qualified enquiries in September, which makes them worth monitoring, even though the volume remains modest.

The report does not establish which Google AI feature produced a particular visit or whether a Bing citation led to a booking. It also does not prove that a content update caused the growth. The owner checks which landing pages received enquiries and compares those findings with the visibility reports.

A short summary for the monthly report

“Overall Google Search impressions rose 20% and clicks rose 7.1%. The included generative AI impressions rose 50%, while Bing recorded 480 citations across its supported AI experiences. GA4 recorded 2,600 Organic Search sessions and 120 AI Assistant sessions. These channels produced 36 and four qualified enquiries respectively. Completed sales associated with their recorded enquiry sources generated £4,200, including jobs originating in earlier months. Next month we will review the service pages receiving enquiries and improve any unclear quote information.”

Choose the next action from the evidence

• If relevant visibility grows but visits remain flat, examine the pages, markets and search presentation. Do not assume all exposure will generate a click.
• If visits grow but completed enquiries do not, test the page and form. Check suitability, pricing information and the next step before commissioning more content.
• If enquiries grow but quality falls, review which services and locations those pages attract. The team’s qualification notes matter more than the raw form count.
• If qualified leads grow but completed sales lag, review follow-up time, availability and the sales cycle. Compare lead cohorts rather than mixing this month’s leads with older sales.

Keep traditional SEO diagnostics in the report

Rankings, click-through rate, engagement and indexing still help explain a problem. Use them as diagnostics alongside business results. There is no universal 2% click-through-rate threshold for deciding that a page needs rewriting.

A customer may get the answer on one page and immediately enquire, so more pages per session is not automatically better. Likewise, an average position across unrelated queries can hide the performance of an important service page. Segment by commercially relevant pages and markets.

When assessing return on investment, compare the profit associated with attributed sales against SEO costs, using a stated method. Revenue alone is not profit, and attributed revenue is not proof of additional sales that would never have happened otherwise.

Make the monthly report repeatable

• Use complete calendar months and retain the report property, filters, channel scope, currency and export date.
• Record tool time zones and allow for data processing. Month boundaries can differ between Search Console and your analytics property.
• Keep missing or unavailable values separate from genuine zeros. Check collection before interpreting an empty report as no activity.
• Keep a record of website changes, campaigns and tracking changes. Compare the same months a year apart when seasonality matters.
• End with the main business finding, a material limit and one or two actions with an owner. A dashboard full of numbers is not an action plan.

Our recommendation

Start with the outcomes your business needs, then use search and analytics reports to explain what is happening along the way. Keep impressions, citations, sessions and sales in separate rows with honest labels.

A small monthly report that leads to a useful decision is more valuable than a single “AI SEO score”. Improve the pages customers need, check that measurement works and assess the quality of the business those pages bring.

Sources and further reading

Google — Generative AI performance reports

Google — Generative AI Search report help

Microsoft — AI Performance public preview

Microsoft — New AI visibility insights

Google Analytics — Product updates

Google Analytics — Default channel definitions

Google Analytics — Traffic acquisition

Documentation reviewed on 11 October 2026. Check the linked guidance as reporting features and classifications evolve.