Click-through rate (CTR): definition, and why AI answers break it

Click-through rate is a ratio with two inputs:

• Impressions: how many times a link to your page was shown.
• Clicks: how many times someone followed that link.

CTR = clicks ÷ impressions × 100. Say a page appeared 1,000 times in search results and earned 30 clicks. Its CTR is 3%.

The formula is simple. The hard part is that both inputs must describe the same link, on the same surface, over the same period. If either input is missing or comes from a different system, the ratio stops meaning anything.

In classic Google Search, one system records both halves. Search Console's performance report shows impressions, clicks, CTR and average position for each query and page. Google counts the impression when it shows the result and counts the click when someone follows it. The denominator and numerator share a definition, so the ratio is trustworthy.

That's why search CTR became a standard diagnostic. A page with high impressions and low CTR usually has a title or snippet problem. A page with low impressions has a visibility problem. You can tell the two apart because both numbers sit side by side.

Next move: in classic search, read CTR per query, not sitewide. A blended CTR hides which pages need work.

AI answer surfaces rarely give you both inputs. Here is what each one exposes, based on vendor documentation reviewed in August 2026.

Google's generative AI report launched in June 2026 with impressions only. Google described more metrics as a future intention, not a feature. You can see whether you appeared, but not whether anyone clicked. Our answer on whether Search Console shows AI Overview clicks covers the details.

ChatGPT has the opposite gap. Some clicks arrive tagged, but OpenAI publishes no impression count to site owners. Retrieval is also discretionary: the model decides per query whether to search the web at all, so you can't infer how often your page was shown.

There's a deeper issue. Being cited isn't the same as being shown prominently or being used. Mechanistic research on retrieval-augmented models found citation output can be changed without changing the sources, so a citation count is a noisy proxy for influence.

Next move: record AI impressions, citations and tagged referral sessions as three separate metrics. Use the AI Overviews traffic guide for a measurement setup that doesn't depend on a ratio you can't compute.

• "My AI Overviews CTR is X%." No first-party source gives you clicks for that surface, so any such number was assembled from mismatched inputs.
• "Analytics clicks over Search Console impressions is close enough." Analytics counts sessions it can attribute. Search Console counts impressions Google chose to report. Dividing one by the other mixes definitions.
• "Falling CTR proves AI is taking my traffic." It might. It might also be a ranking change, a seasonal shift or a reporting bug. Google confirmed a bug that depressed impressions in its performance reports from around mid-August 2026, so check the window before drawing conclusions.
• "A citation means a click is coming." Many cited pages get little traffic. See why AI can cite your site but send almost no traffic.
• "There's an AI position to pair with CTR." Google's AI report exposes no position data, so no AI rank exists to check against.

• Zero-click search: searches that end on the results page, which shrink the click side of CTR.
• Citation rate: how often an AI answer cites your pages, a metric you can actually measure.
• Citation rate vs citation share: two AI metrics that are easy to confuse.
• Mention rate: how often an AI answer names your brand, cited or not.

What is a good search CTR? There is no universal benchmark in the evidence we rely on. CTR depends on the query, the position, the snippet and how many features crowd the results page. The useful comparison is your own pages against each other for similar queries, and the same page before and after a change. Treat any published average as context, not a target.

Will Google add clicks to the generative AI report? Google described additional metrics only as a future intention when it announced the report in June 2026. Until a click metric actually appears in the report, plan your measurement around impressions. Check the report itself rather than relying on older help pages, because Google's documentation and its rollout have not always matched.

Can I measure ChatGPT traffic at all? Partly. ChatGPT adds utm_source=chatgpt.com to many referral links, so your analytics can attribute those sessions. The tagging is incomplete, with free-tier and mobile-app clicks often arriving untagged. Treat tagged sessions as a floor, never as a full count, and don't divide them by anything to make a CTR.

Should I stop tracking CTR? No. CTR stays a reliable metric for classic search, where one system records both impressions and clicks. Keep using it there. For AI surfaces, report what each platform actually exposes: impressions in Search Console, citations in Bing Webmaster Tools, and tagged referral sessions in analytics. Keep each in its own column.