Neither AI Overviews nor traditional blue links are a universal conversion winner. The answer depends on the query, the search interface shown, the user's location and device, the page selected, and what the business counts as a conversion. A defensible comparison therefore separates exposure, visits, conversion rate and total conversions instead of turning them into one headline.
Google states that traffic from AI features is included in Search Console's overall Web search reporting. In June 2026, Google also announced a dedicated Gen AI performance report rolling out to an initial subset of sites. Availability is not universal, so the measurement plan must work whether or not a property has that report. See Google's AI-search optimization guidance and the Gen AI performance report announcement.
What each metric can answer
| Metric | Question it answers | What it does not prove |
|---|---|---|
| Impressions | How often a measured search result or supported AI feature was shown under the reporting definition. | That the user noticed the brand or read the answer. |
| Clicks or sessions | How many measured visits reached the site. | That the visit was incremental or caused by one interface element. |
| Conversion rate | What share of measured visits completed a defined event. | Which channel produced more total conversions. |
| Total conversions | How many measured outcomes were completed. | That every outcome was caused by the immediately preceding click. |
| Assisted or later conversion | Whether an observed touchpoint preceded a conversion in the selected attribution model. | True causal incrementality. |
A smaller, high-intent visit cohort can have a higher conversion rate while still producing fewer conversions. A large blue-link cohort can produce more total conversions with a lower rate. Report both the numerator and denominator.
Build a measurement table before interpreting results
Start with one row per landing page, query cohort, country, device and week. Preserve those dimensions because a blended total can hide a material change in query mix. Include Search Console impressions and clicks, analytics sessions, the agreed conversion event, conversion count and revenue or qualified-pipeline value when available.
If the dedicated Gen AI report is available, keep its rows separate from the Web report and record the report's documented scope. If it is not available, do not manufacture an AI Overview segment from screenshots or page position alone. Use the data Google actually exposes, then label manual interface observations as observations rather than traffic attribution.
Use stable conversion definitions
Choose the business outcome before comparing cohorts. A newsletter signup, demo request, completed purchase and qualified opportunity are not interchangeable. Document:
- the event name and analytics definition;
- the first date the event was implemented consistently;
- consent or tracking conditions that can reduce observed sessions;
- the attribution window and model;
- known bot, internal and duplicate-event exclusions;
- any form, checkout or CRM change during the comparison period.
Validate the event in the browser and destination system. A conversion-rate change is meaningless when the event fired twice, stopped firing on one template or was renamed halfway through the period.
Compare matched cohorts, not unlike totals
A useful comparison holds as much context stable as possible. Compare the same landing-page group, topic or intent, market, device class and time window. Annotate major site releases, promotions, outages, pricing changes and search-interface changes. For low-volume cohorts, show raw counts and avoid declaring a winner from a handful of conversions.
Seasonality also matters. A year-over-year comparison can be more informative than a week-over-week comparison, but only when tracking and page purpose are comparable. If they are not, say so.
Calculate rate and volume separately
For each cohort, calculate:
- Click-through rate: measured clicks divided by measured impressions under the report's definition.
- Session conversion rate: measured conversions divided by eligible sessions.
- Conversion volume: the raw number of conversions.
- Value per session: measured revenue or qualified value divided by eligible sessions, when the value is reliable.
Do not compare a Search Console click denominator with an analytics-session denominator without explaining the difference. Clicks and sessions can diverge because of consent, repeated visits, redirects, browser behavior and implementation details.
Review landing-page quality as a separate layer
An AI feature may influence the type of visitor who clicks, but the landing page still determines whether the user can complete the next step. Review message match, page speed, mobile usability, evidence, pricing or eligibility clarity, form friction and error states. A conversion problem on one landing page is not proof that an entire search interface produces weak traffic.
Use the AI SEO 101 guide for the technical and content baseline. For query-level visibility observations, use the KNWN measurement guide.
Interpret changes without claiming causation
A before-and-after change is evidence of correlation. Search demand, competing results, product changes and measurement changes may move at the same time. When the decision justifies the effort, use a controlled experiment, a matched holdout or a phased rollout. Otherwise report the observed change and its limits plainly.
A defensible conclusion sounds like this: “For the 42 non-brand landing pages in this market, measured organic sessions fell 8% while qualified-form conversion rate rose from 2.1% to 2.6%. Total qualified forms were unchanged. The property did not have Google's dedicated Gen AI report, so this analysis cannot attribute the movement specifically to AI Overviews.”
Monthly review checklist
- Confirm the reporting scope and feature availability.
- Validate conversion events and CRM reconciliation.
- Export the same query, page, country and device cohorts.
- Report impressions, visits, rate and volume separately.
- Annotate releases and external changes.
- Investigate landing-page differences.
- State what the evidence can and cannot support.
The objective is not to crown one interface. It is to understand where demand is visible, which visits arrive, what those visitors do and where the business should improve the experience.
Interpretation note
This article is educational material, not a product commitment or a guarantee of rankings, citations, traffic or commercial outcomes.