What Do You Do With a Page AI Cites but Nobody Visits?

What do you do with a page AI cites but nobody visits?

You decide on purpose whether that page exists to earn traffic or to earn a mention, and then you stop measuring it the wrong way. Most sites have a handful of pages that answer questions so completely that an AI engine quotes them and the reader never needs to arrive. That is not a failure. It is a job description.

The reason this feels uncomfortable is that the whole industry built its sense of whether content works on the click. A page with impressions and no sessions used to mean something was broken. Now it can mean the opposite, that the page is so clearly written that a machine could lift the answer straight out of it.

The problem is that both situations look identical in analytics. You cannot tell a page that was useful from a page that was ignored by staring at a sessions column. So the work is to separate the two on purpose rather than hope the numbers sort themselves out.

Why does a cited page get so few clicks?

Because the answer is already on the screen. Pew Research Center found that users who encountered an AI summary clicked on a traditional search result link in 8 percent of all visits, while those who did not encounter an AI summary clicked on a search result nearly twice as often, in 15 percent of visits. The summary satisfies the question.

Those figures come from Pew's browsing study of 900 US adults who are members of KnowledgePanel Digital, covering March 2025 and 68,879 unique Google searches. Pew also reports that 18 percent of all the Google searches in that study generated an AI summary. So this is not yet the majority experience, but it is a large minority, and it behaves very differently.

The shape of the summary explains a lot of the behaviour. Pew found the typical AI summary in their study was 67 words long, and that the vast majority of them, 88 percent, cited three or more sources. Sixty-seven words is a complete answer to most simple questions. Three or more sources means your citation is sharing the credit even when you win.

What I take from that combination is that citation and click are drifting apart as outcomes. A short answer with several sources is an answer designed to end the search, not to start a journey. If your page is one of the three, you were useful and you were not visited, and those two facts are now compatible in a way they never used to be.

Is a citation worth anything on its own?

Yes, but less than a click and more than nothing. Be sceptical of anybody claiming to know the exact exchange rate. A citation puts your name in front of someone at the moment they were deciding something. That has value. It is also hard to measure with ordinary tools, so treat it as a leading indicator rather than revenue.

The honest position is that nobody outside the engines has clean data connecting citations to outcomes. I have seen confident numbers put on this and I do not trust any of them, because the measurement problem is genuinely hard and the incentive to sell a number is genuinely strong. What you can observe is direction: are you cited more this quarter than last, on the questions that matter to your business?

There is one signal worth noting from Pew's data that points at where citations come from. They report that 6 percent of the sources linked in AI overviews were to .gov websites, compared with just 2 percent for standard search results. Whatever else is happening, the mix of sources in a summary is not the same as the mix in a results page, which tells you the selection is doing something different.

How do you even find these pages?

Compare impressions to clicks at the page level and sort by the widest gap. Pages with strong visibility and almost no sessions are your candidates. Then check each one by hand against the questions it answers, because the gap alone does not tell you whether you were quoted or simply skipped.

The manual check is unavoidable and it does not take long. Take your top twenty candidate pages, pick the main question each one answers, and ask that question of the engines your buyers actually use. If your page appears as a source, you have a citation-only page. If it does not, you have a visibility problem of the ordinary kind.

Keep the results in a plain spreadsheet with a date column, and repeat it quarterly. This is deliberately low-tech. Citation tracking tools exist and some of them are useful, but starting with twenty manual checks teaches you what your pages are actually being used for, which no dashboard will tell you. I made a similar argument about the structure that gets a page quoted in the piece on why AI search quotes one paragraph.

Should you rewrite a cited page to pull the click back?

Usually no. Making your page harder to quote does not make people visit it, it makes the engine quote somebody else. You cannot force a click by withholding an answer, because there are always three sources and one of them will answer plainly. Withholding just removes you from the summary.

What you can do is change what a reader would gain by arriving. If the summary can carry your whole answer, the page needs something the summary structurally cannot carry: a tool, a template, a calculator, a decision the reader has to make with their own inputs, or depth that does not compress into 67 words. The click has to be earned with something uncopyable, not protected by being unhelpful.

The version of this I see most often, and dislike most, is the page that answers the question in paragraph nine after eight paragraphs of throat clearing. That does not preserve the click. It just means the engine extracts paragraph nine and the human who did arrive leaves annoyed. Answer first, then add the thing the answer cannot contain. I laid out the mechanics of that structure in the article on writing answer blocks.

Which pages deserve to be citation-only?

Definitional and factual pages, mostly. What a term means, how a process works, what a standard requires, which format a file needs. These questions have complete short answers, and fighting for the click on them is a losing position. Let them do their job as evidence that you know the field.

The pages that should never be citation-only are the ones where the decision is the product. A comparison between two options, a pricing explanation, a case study, a page where somebody is choosing whether to work with you. If those are being fully summarised, something is wrong with how they are written, because a real decision does not fit in 67 words and should not appear to.

Sorting your library into those two buckets is a useful afternoon. It changes what success looks like per page, which changes what you brief writers to do. A glossary entry and a services page have completely different jobs in an AI-mediated search world, and briefing them the same way is why so much content underperforms at both.

What does this do to how you value content?

It forces you to value the portfolio rather than the page. Some pages will earn citations and no traffic, some will earn traffic, and a few will earn both. Judging every page by sessions will lead you to delete the ones doing the reputational work and keep the ones doing nothing but ranking.

I have started thinking about it as two different products made of the same words. One product is a destination for a human being. The other is source material that machines summarise. A single page can be both, but it will be better at one, and pretending otherwise produces content that is mediocre in both directions.

This also changes what a content audit looks like. The old audit asked which pages get traffic and which should be cut. A current audit has to ask which pages are being used by machines, because a page with no sessions may be the most quoted asset you own. Cutting it because it is quiet is an expensive mistake that shows up nowhere in your analytics.

How do you report this to someone who only counts sessions?

Show the gap and name it. Put impressions next to sessions for the pages in question, add the citation checks you ran by hand, and label those pages as citation assets rather than traffic assets. The goal is a shared vocabulary, not a new metric. Once the label exists, the conversation stops being an argument.

Be careful not to oversell. If you tell a founder that citations are the new traffic and then cannot connect them to anything, you will burn credibility you need later. The honest framing is that this is a measurable change in how search behaves, that Pew's numbers show clicks falling when a summary appears, and that you are tracking your position in the summaries while the measurement matures.

I also think it is worth putting the zero-click shift in front of leadership as a trend rather than an incident, which is the case I made in the article on zero-click search for site owners. A trend gets planned for. An incident gets panicked about, and panic produces exactly the wrong response, which is usually to publish more pages faster.

What is the risk of chasing citations?

That you optimise for being quoted and forget you are running a business. A citation does not sign a contract. If your content strategy becomes a race to be the three sources in every summary, you will produce a library of correct, quotable, forgettable pages and no reason for anybody to choose you over the other two sources.

The deeper risk is that citations are an easier scoreboard than revenue, and easier scoreboards always win the attention of busy teams. I would rather have five pages that convert and no citations than fifty citations and no pipeline. The moment the citation count becomes the goal, you have swapped a hard question for a comfortable one.

My rule is that citation work earns budget only where the audience is genuinely in a research phase. If the question a page answers is one somebody asks while deciding whether to buy, the page should be built to convert. If it is one they ask while learning the field, let it be quotable and be pleased when it is quoted.

What should you do next?

Pull your twenty widest impression-to-session gaps this week and check each one against the engines your buyers use. Label every page as a traffic asset or a citation asset. Then fix only the pages where the label and the page's actual job disagree, which is usually about three of them.

After that, change one thing in your briefs. Tell writers which of the two jobs each new page is for, before they start. That single line prevents most of the mismatch, and it costs nothing compared to rewriting a library later once the gaps have already opened up.

If you are looking at a set of pages with good visibility and no visits and cannot tell whether that is success or failure, reach out. Working through that distinction for a real site is one of my favourite parts of this job, and it usually takes one honest afternoon rather than a project.