Every month we have the same conversation with a client: rankings are holding, the content is performing, the business is getting more enquiries that mention things only the website says — and organic sessions are flat or down. Nothing is broken. The measurement is.
With AI-generated answers now the default experience for the majority of searchers, a growing share of your SEO work pays off somewhere your analytics never sees: inside an answer panel, inside a chat thread, inside a recommendation someone acts on three days later. If your reporting still treats organic sessions as the score, you are grading the wrong exam.
Three things happen at once. Answers resolve the query on the results page, so the click never happens. Assistants like ChatGPT, Perplexity, Gemini and Copilot cite you inside a conversation, and many of those visits arrive stripped of referrer data — landing in your reports as "direct". And the research phase now stretches across several tools before anyone types your brand name, so the visit that finally converts looks like it came from nowhere.
The result is a systematic undercount. The work is producing demand; the dashboard is measuring one narrow expression of it.
Before adding anything exotic, capture the signals already flowing through your own systems.
In GA4, build a custom channel group that matches assistant domains — chatgpt.com, perplexity.ai, gemini.google.com, copilot.microsoft.com and their variants — so those sessions stop hiding inside "Referral" and "Direct". Volume will look small. Watch the conversion rate instead: these visitors usually arrive pre-qualified, having already read a summary of what you do.
Your access logs record which AI crawlers fetch which URLs and how often. If GPTBot, ClaudeBot, PerplexityBot and Google's crawlers are hitting your service pages weekly, your content is in the retrieval pool. If they never reach a section of the site, that section cannot be cited — and that is a fixable technical problem, not a content one.
In Search Console, rising impressions with flat clicks on informational queries is the fingerprint of answer absorption. It is not a failure — it means you are being surfaced and summarised. Track it deliberately rather than reading it as decline.
The prompt-set audit is the single most useful new habit. Pick the twenty questions a real buyer would ask before hiring you, run them monthly, and log whether your brand appears and how it is described. That gives you a citation rate you can trend — the AI-era equivalent of a rank tracker.
Pair it with branded search volume, which is the cleanest proxy for demand created by visibility you cannot click-count. When unbranded sessions dip but branded searches and direct traffic climb, the strategy is working; people are finding you elsewhere and coming back by name.
Traffic was always a proxy for demand. When the proxy breaks, measure the demand.
Lead the report with business outcomes — enquiries, qualified leads, revenue — then show visibility metrics as the supporting layer: citation rate, impressions, crawler coverage, branded search, AI referral quality. Sessions stay in the report, but demoted to one input among several rather than the headline.
Set the expectation early with whoever signs off the budget. A quarter into a shift like this, the honest sentence is: "fewer people are clicking, more people are arriving already convinced." That is only a bad story if nobody explained it in advance.
You cannot fix what you refuse to measure, and you cannot measure this the way you measured 2019. Instrument the referrals you can still catch, audit your citations on a fixed prompt set, watch branded demand, and report against leads rather than sessions. The channel did not stop working — the scoreboard did.
We build SEO measurement that tracks AI citations, branded demand and real enquiries — not just the sessions your analytics happens to catch.
Talk to us