Short answer, as of September 2026: expect citation without clicks to be the normal case, and split your content accordingly — some pages exist to be quoted and build preference, others exist to be visited and convert. Treating every page as a traffic acquisition asset is what makes AI search look like a threat rather than a channel.
This is the strategic shift most teams have not made yet. The measurement side of it — why AI referrals barely register in analytics and what to instrument instead — is covered in why AI citations barely show up in Google Analytics. This piece is about what to do about it.
The mechanic, stated plainly
When an assistant answers a question using your page, three things can happen. The user gets what they needed and stops. The user clicks a citation link and lands on you. Or the user does not click now, but remembers a name, and arrives later by searching for you directly.
The first outcome is the most common for informational questions, and it will not show up in your analytics as anything at all. The third is valuable and effectively invisible in attribution, which is why it gets systematically undercounted in decisions about what to publish.
If your internal model is "content earns sessions, sessions earn conversions", the first and third outcomes both read as zero. That model was already strained by search results answering questions on the page; assistants finish the job.
Split the content, do not fight the mechanic
The practical response is to stop treating your content as one pool with one job. Two jobs, different success criteria.
Quoted content is written to be the passage an assistant leans on. Definitions, comparisons, cost explanations, process descriptions, "what happens if" questions, limitations. It succeeds when your business is named and characterised accurately, whether or not anyone visits. Its metric is citation and share of voice, not sessions.
Visited content requires the user to be on your site for the value to exist. Tools and calculators, configurators, quote forms, account and pricing flows, anything interactive, anything where the answer depends on the user's own inputs. An assistant cannot substitute for these. It can, at best, tell somebody they exist and where.
Most teams have plenty of the first and too little of the second, which is exactly backwards for the environment. The single highest-leverage response to zero-click is to build more things that genuinely cannot be answered in a paragraph.
| Content type | Can an assistant fully substitute it? | Success measure | Strategic role |
|---|---|---|---|
| Definitions and explainers | Yes, almost always | Citation, accurate characterisation | Preference and trust |
| Cost and pricing explanations | Yes, if you publish numbers | Citation, being named on cost prompts | Qualification, preference |
| Comparisons including your limits | Yes | Being named as the right or wrong fit | Qualification |
| Calculators and configurators | No | Sessions, completions | Acquisition and conversion |
| Quote and booking flows | No | Conversions | Conversion |
| Proprietary data and original research | Partly, but it must cite you | Citation plus referral | Authority, link acquisition |
| Account and support tooling | No | Retention | Retention |
Original research sits in an interesting position: assistants will summarise it, but because the figure belongs to you, the citation is nearly obligatory and your name travels with it. That makes it one of the few content types where being quoted is unambiguously good even when nobody clicks.
What being named without being visited is actually worth
It is not nothing, and it is not a click. It is closer to a recommendation.
Three things happen when an assistant names you. The user learns you exist in a context where you were presented as a credible answer rather than an advertisement. They may later search your brand directly, which is a measurable event even though its cause is not. And if they are comparing, being present in the consideration set is the difference between being evaluated and not.
The practical consequence is that brand-name search volume, direct traffic and "how did you hear about us" answers become part of how you read AI search performance. None is clean. Together they are considerably better than the referral number, which is close to useless on its own. The case for framing this to a sceptical stakeholder is made in proving the business value of AI search.
Publish the numbers, or be characterised without them
The counter-intuitive move that pays off here is publishing specifics you previously withheld — particularly prices.
The old logic was that vague pricing brought people to a form. In an environment where an assistant answers the cost question regardless, withholding the number does not prevent the answer; it just means the answer is assembled from competitors, directories and forums, none of which are trying to represent you accurately. You have not protected the information. You have outsourced it.
Publishing a real range makes you the source of your own pricing, filters out buyers who were never going to proceed, and gives the assistant something specific to quote. The commercial objections are real and are worked through in publishing prices and cost transparency.
Make the citation work harder
If most users will never arrive, the citation itself has to carry more.
That means the passage an assistant is likely to quote should contain the things you want a non-visitor to know: who you are, what you do, the qualifying constraint, and the distinguishing specific. A paragraph that answers the question but never identifies the business behind it gets used and gives you nothing.
This is not an argument for stuffing your brand name into every paragraph, which reads badly and gets the passage skipped. It is an argument for making sure the answer-bearing passages are self-contained enough that quoting one tells the reader who answered. Retrofitting existing pages this way is usually faster than writing new ones; the method is in retrofitting existing pages for citation.
What not to do
Do not gate everything. Putting your explanatory content behind a form removes you from the answer entirely. The competitor who published theirs becomes the source, and the assistant characterises your category using their framing.
Do not conclude that content stopped working. Sessions falling while brand searches and qualified enquiries hold steady is a channel shift, not a failure. Cutting content budget on the referral number alone is the predictable and expensive mistake of this period.
Do not chase clicks with worse content. Deliberately withholding the answer to force a visit is the tactic assistants are best at routing around. It produces pages that are not quoted and not visited.
A reasonable position to hold
Publish the explanatory content and accept that most of its value arrives as preference rather than sessions. Build the interactive things that cannot be summarised, and treat those as your traffic and conversion assets. Measure citation and brand demand alongside sessions, and stop asking one number to represent both.
The underlying capability work does not change — being retrievable, specific and corroborated is what gets you quoted in the first place, as set out in how to get cited by ChatGPT and Perplexity. What changes is what you expect in return, and whether your reporting is built to see it. You can see what the tracker records on the features page.
Frequently asked questions
Is it worth being cited by AI if nobody clicks through? Yes, though the value arrives as preference rather than sessions. Being named presents you as a credible answer rather than an advertisement, puts you in the consideration set for comparisons, and drives later brand searches that show up as direct or branded traffic instead of referrals.
How do I stop AI search from reducing my website traffic? You largely cannot stop it for informational content, and trying to by withholding answers produces pages that are neither quoted nor visited. The effective response is to build more content that genuinely cannot be summarised, such as calculators, configurators and quote flows.
Should I gate my content to force people onto my site? No. Gating removes you from the answer entirely, and the competitor who published theirs becomes the source the assistant uses to characterise your whole category. You lose the citation and rarely gain the form fill.
What content still earns clicks in AI search? Anything whose value depends on the user's own inputs or on being on your site: calculators, configurators, quote and booking flows, account tooling, and interactive comparisons. An assistant can mention these exist but cannot substitute for them.
How should I measure AI search if referrals are unreliable? Read citation rate and share of voice alongside branded search volume, direct traffic and what customers say when asked how they found you. No single one is clean, but together they are far better than a referral number that captures only a fraction of the effect.
Does publishing prices make the zero-click problem worse? No, it usually improves your position. The cost question gets answered whether or not you participate, so withholding the number means the answer is assembled from competitors and directories rather than from you. Publishing a real range makes you the source of your own pricing.
My traffic is down but enquiries are steady. What does that mean? Usually a channel shift rather than a content failure. Some users are getting the informational part of their journey answered off-site and arriving later with more intent. Check branded search and direct traffic before cutting content investment on the sessions number alone.
Should I write differently if most readers will never visit? Yes, in one specific way: make the answer-bearing passages self-contained enough that quoting one identifies who answered it. A paragraph that answers the question but never establishes the business behind it gets used and returns nothing.