Measure AI search traffic in 2027 without overstating revenue

Use external research as context, not your conversion forecast

Pew’s 2025 study examined 68,879 unique searches from 900 US adults. Traditional-result links were clicked on 8% of visits with AI summaries and 15% without; summary-source links were clicked on 1% of visits with summaries. Those findings are not your website’s conversion rates. Pew: clicks on Google pages with AI summaries.

Its browsing period was March 2025, while results were re-collected in April. Answers can change between observations. Use the research to justify checking your own acquisition mix, not to claim that every summary appearance loses the same fraction of clicks. Pew: clicks on Google pages with AI summaries.

Give each measurement layer its own denominator

For answer observations, report the number of completed captures, the question set, the engine surfaces, and the capture dates. Count a visible mention only within those completed captures. Do not describe a handpicked set of screenshots as the share of all buyers seeing your brand.

For visits, record the analytics source definition and the landing page. Google Analytics distinguishes user acquisition from session-based traffic acquisition; switching their scopes can change what a comparison means. Pick the scope that answers your question and keep it consistent. Google Analytics: user versus traffic acquisition.

Keep direct traffic and unidentified visits honest

Google Analytics describes direct/none as traffic without a clear referral source. It can result from several causes. Do not relabel a rise in direct traffic as AI traffic because an AI campaign happened at the same time. Google Analytics: direct/none traffic.

Use campaign tags on links you control, preserve them across redirects, and document the referral rules used in your analysis. For links you do not control, accept that some attribution will remain missing. Customer-reported discovery can add context, but it is a different evidence source from a recorded session.

Connect qualified outcomes before estimating commercial value

Define the next useful business event: a qualified audit request, a relevant demo request, or another action with an accountable owner. Verify that the event represents a real successful action rather than a button click. An event firing twice or on a failed form can create a convincing but incorrect growth story.

Inspect a small sample of records manually. Can you identify which landing page and source are recorded, whether the enquiry meets the business definition, and whether a later outcome is known? If the chain breaks, report the break rather than filling it with an assumed conversion rate.

Compare changes with their alternative explanations

When you update source pages, record the release date, the affected question group, pricing changes, campaigns, and provider changes. Compare a stable baseline with subsequent periods, but avoid attributing the entire difference to your edit. A comparison group can strengthen the design when one is practical; it does not erase every source of bias.

End each review with a decision: continue a useful content fix, investigate an acquisition gap, or improve instrumentation. A good measurement plan helps allocate effort even when causal revenue attribution is unavailable. It should make uncertainty actionable rather than hide it behind a larger number.

Questions answered