A Pattern That Shouldn’t Be Possible
Open Search Console and sort your queries by impressions. Look for anything sitting at an average position of 1.0 or 1.1. Now look at the click column next to it.
If you see hundreds of impressions at position one and a click count of zero, you’ve found the thing this article is about. We found it in our own Search Console data this spring, on several commercial queries — ranking first, week after week, with a click-through rate that rounded to nothing.
Position one should convert at somewhere between 25 and 40 percent. Zero is not a bad result. Zero is a signal that the word “position” no longer means what it used to.
What Those Impressions Actually Are
Search Console reports an impression when your page appears in a result set. It does not distinguish between the classic ten blue links and the newer surfaces Google has been bolting onto the results page. So an impression at “position 1” can now mean any of the following:
- A citation inside an AI Overview. Google answers the question at the top of the page and lists sources beside it. You’re the source. The user has their answer and no reason to click.
- A grounding fetch from AI Mode or an assistant. When an AI product runs a search to ground its answer, that search generates impressions. Nobody was ever going to click, because no human was looking at a results page.
- An appearance in a SERP feature — a knowledge panel, a carousel, a “things to know” accordion — that renders your content in place.
The tell for the second category is unmistakable once you know it. Go into your query report and look for entries written in full sentences with instructions attached. We have queries in ours that read like this:
“compare wordpress with hubspot, squarespace, and wix for hosting performance and speed. you must provide a forced ranking from best to worst.”
No person types that into a search box. That is a language model decomposing a user’s question into sub-queries and searching each one. We rank in the top three for a cluster of them. Those impressions represent an AI system reading our content to build an answer for someone who will never see our domain name.
This Is Not Necessarily Bad News
The instinct is to treat zero-click impressions as lost traffic. That’s the wrong frame, and it leads to bad decisions.
Being cited in an AI answer is not a failed visit. It’s a different kind of win, with a different payoff curve. When ChatGPT tells someone “a few agencies that specialize in this include…” and your name is in the list, that person did not visit your site. They may still call you next month, or search your brand name directly, or mention you to a colleague.
The measurable version of that shows up somewhere else in your data: branded search volume. Our own branded queries — the name of the company, plus every misspelling of it — carry a click-through rate several multiples higher than anything else we rank for. That’s the pattern to watch. Unbranded impressions go up, unbranded clicks stay flat, branded searches climb. The traffic arrives, just through a different door and on a delay.
The honest caveat: this is hard to attribute cleanly, and anyone who tells you they can measure AI citation ROI precisely is selling something. What you can do is stop reading zero-click impressions as failure.
What Actually Changes About Your Strategy
If a meaningful share of your visibility is now being consumed by machines that summarize rather than humans who click, a few things follow.
1. The click-through-rate lever moves
Title and description rewrites still matter enormously — for the queries where humans are still choosing between links. They do nothing for AI Overview citations. So segment before you optimize. Find the queries where your CTR is below the position-adjusted benchmark and the query looks like something a person typed. Those are your real CTR opportunities. The rest are a different game.
2. Being quotable beats being comprehensive
A language model building an answer pulls the passage that most directly and cleanly states a fact. That rewards a specific writing shape:
- Answer the question in the first sentence under the heading, then elaborate. Don’t build to a conclusion.
- Use headings that are the question, phrased the way someone would ask it.
- Put facts in structures — lists, tables, definition pairs. Prose buries data; structure surfaces it.
- Include the numbers and the caveats. Hedged, specific claims get quoted. Vague superlatives get skipped.
None of this is new advice, exactly. It’s the same advice featured snippets rewarded. It just matters more now, and the payoff is less visible.
3. Entity clarity does heavy lifting
For an AI system to recommend you, it has to be confident about what you are, where you operate, and what you do. That confidence is built from structured data, consistent information across the web, and unambiguous language on your own pages.
Concretely: a connected schema graph where your organization, your pages, and your authors reference each other by ID. Consistent name, address, and phone everywhere they appear. An llms.txt file that states plainly what you do and links your key pages. We covered the mechanics in how to structure your website so AI can recommend it.
4. Your measurement has to change or it will mislead you
If your monthly report is “organic sessions, up or down,” it is going to tell you a story that gets less true every quarter. Add these:
- Impressions split by branded and unbranded. Rising unbranded impressions with flat clicks is the signature of AI consumption, not decline.
- Branded search volume over time. This is the closest thing to an AI visibility proxy you can get for free.
- Referral traffic from AI hosts — chatgpt.com, perplexity.ai, and friends. The volume is small. The intent is very high.
- Direct traffic to deep pages. Someone landing straight on a service page without a referrer often heard about you somewhere a referrer doesn’t survive.
What We’d Do First
If you want to act on this in an afternoon rather than a quarter, do these three things in order.
Pull a 6-month query export from Search Console. Filter to queries with more than 50 impressions and an average position under 5. Sort by CTR ascending. The top of that list is either a snippet problem you can fix today or an AI surface you can’t — and reading the queries tells you which.
Read the actual query strings. Look for full sentences, instructions, comparison prompts, anything that doesn’t sound like a person at a keyboard. That’s your AI grounding footprint, and its size will probably surprise you.
Fix the human queries and leave the machine ones alone. Rewrite titles and descriptions where a person is choosing. For the machine queries, the work isn’t the snippet — it’s making the page easier to quote correctly.
The Uncomfortable Part
Some of this traffic is not coming back. Informational queries that used to send visits now get answered on the results page, and no amount of optimization changes that. The businesses that adapt are the ones that stop measuring the visit and start measuring whether they’re the name that gets said.
That’s a harder thing to put in a dashboard. It’s also, increasingly, the thing that matters.
If you want a read on how visible your business currently is to AI search tools — and what’s making you hard to cite — that’s what our AI search optimization work covers.
