We respect your privacy.

We use strictly necessary cookies to keep you signed in and to protect against CSRF. With your permission we also use a small amount of first-party analytics to improve the product. We do not sell your data and we do not use third-party advertising trackers. See our cookie policy and privacy policy .

← All posts

Shallow pages lose in AI search first

Crawlmind Engineering··5 min read

A shallow page is a page whose whole job is to restate a fact that an AI answer can now give directly, with nothing a reader could not get from the answer itself.

Google's head of Search, Liz Reid, put a name to this risk on the FOMO Sapiens podcast, in an episode published on November 19, 2026. Asked how publishers survive AI search, she described "less interest in sort of very shallow surface level reciting," according to PPC Land's write-up of the interview. Her example was a page built only to tell you the time zone in Japan. Her test was whether a person would want to read the page at all, or whether it exists only as a route to something else.

The same write-up notes that Reid gave no traffic, click, or revenue figures. So treat this as a statement of direction from the person who runs Search, not as a measured effect. It is still worth acting on, because it lines up with what Google has written down for years and with independent click data.

#What Google has already said in writing

None of this is new policy. Google's guide to creating helpful, reliable, people-first content asks whether content is primarily made to attract visits from search engines, and calls a yes to that kind of question a warning sign. It also asks whether you have an audience that would find the content useful if they came to your site directly.

When AI Overviews and AI Mode expanded, Google's Search Central post on succeeding in AI experiences on Search repeated the same core advice: focus on unique, satisfying content for people. Google's documentation on AI features and your website says the same foundational SEO best practices apply to AI features, and that there are no additional requirements to appear in AI Overviews or AI Mode.

Reid's comment adds one thing the documents do not say as plainly. A page that only recites a fact now competes with the answer box itself, and the answer box usually wins.

#Why recitation pages are the first to go

The mechanism is simple. A user who asks for the time zone in Japan gets the answer at the top of the result. The page that held that fact has nothing left to offer the user, so there is no reason to click.

Independent data shows the size of the click shift when AI answers sit in front of links. In a preregistered field experiment with about 1,100 US Google users, people assigned to AI Mode clicked through to outside websites 18.8 percentage points less often than people on standard Google, as reported by PPC Land. The experiment ran for seven days and the sample skewed younger and more educated, so the exact figure should not be generalized. The direction is clear enough.

That study measured all clicks, not only clicks to shallow pages. Reid's point suggests where the loss concentrates: on pages where the answer is the entire value. A page with original data, a worked method, or first-hand testing still has something the summary cannot carry in full.

#How to find your shallow pages

You do not need a new tool to start. Use your own analytics and a short checklist.

  1. List pages that answer one closed question. Think "what is X", "how many Y", "when does Z open", unit conversions, glossary entries, and single-fact FAQ pages. These are recitation candidates by default.
  2. Check whether the answer fits in one sentence. If the entire useful content of the page can be stated in one sentence, an AI answer can state it too.
  3. Look for anything only you can say. Original numbers, screenshots from your own product, a method you tested, a decision you made and why. If none of these exist on the page, it is shallow by Reid's definition.
  4. Compare impressions with clicks over time. In Search Console, pages whose impressions hold steady while clicks fall are the ones where the answer is being served without the visit. Sort by that gap.
  5. Ask Reid's question directly. Would a person who found this page through a friend's link read past the first paragraph? If not, the page exists for the search engine, which is the pattern Google's guide flags.

#What to do with each one

Every shallow page falls into one of three groups.

Merge it. Many sites have dozens of thin pages that each answer one narrow question on the same topic. Fold them into one reference page that answers the questions together and adds the context that connects them. Redirect the old URLs to the new section.

Deepen it. If the topic matters to your business, add what the summary cannot give: your own measurements, a worked example with real inputs, the edge cases you have seen customers hit, or a comparison you ran yourself. The atomic answer can stay at the top. It just stops being the only thing on the page.

Retire it. Some pages were built only to rank for a lookup query and have no audience outside search. If they have no links, no conversions, and no clear path to becoming useful, removing them is a reasonable call. Mark the URL as gone, or redirect it to a relevant parent page.

Avoid the fourth option many teams reach for, which is padding. Adding several paragraphs of generic background to a one-fact page does not make it deeper. It makes it a longer shallow page, and Google's guide specifically warns against writing to a target word count.

#Keep the answer, add the reason to stay

The answer-first structure still matters. AI systems need a clear, extractable statement to cite, and readers want the answer without scrolling. The change is what follows that first sentence. A page that leads with the answer and then gives evidence, method, and judgment can be cited and still earn the click. A page that leads with the answer and then ends has handed its value to the summary.

Reid's framing also gives a useful editorial rule that has nothing to do with algorithms. Before you publish a page, write down who would want to read it if search did not exist. If the only honest answer is "a crawler", the page is already at risk.

#Sources

Related field notes

Share or discuss

Field notes in your inbox

New posts, no spam. Roughly monthly. Unsubscribe with one click.