What AI Overviews do to your traffic, and how to tell

Third-party studies report lower click-through on queries that show an overview. Google says it sends higher-quality clicks. Both can be true, and neither is a figure for your site.

1. 1

   Open the Performance report and compare periods

   Compare the last three months to the previous three, or to the same period last year if your traffic is seasonal. Use the Queries tab.

2. 2

   Sort by the difference in clicks

   You are looking for queries where clicks fell and impressions did not. That shape means Google still shows you as often, and fewer people are clicking through.

3. 3

   Split informational from commercial

   Group the falling queries by what the searcher wanted. Questions, definitions and how-tos on one side; product, pricing, comparison and brand queries on the other.

4. 4

   Check the results page for the worst few

   Search each one in a clean window. Note whether an overview opens, what it says, and which pages it cites.

5. 5

   Write down what you found, with the date

   You are building a baseline. Next quarter's comparison is only useful if you know what the page looked like this quarter.

Other things produce the same shape, so rule them out before blaming the overview: a title you changed, a rival who now outranks you, a features change on the results page, or seasonality.

An overview appears on a subset of queries, mostly questions and informational ones, and cites a handful of pages. That tells you where the risk sits and where it does not.

• Most exposed: definitions, simple how-tos, and anything answerable in three sentences. If the answer fits in the box, the box is enough for many searchers.
• Partly exposed: comparisons and best-of queries. The overview summarises, and people still click to check the detail and the prices.
• Barely exposed: your brand name, pricing, login, docs and anything where the searcher wants your page specifically.
• Not exposed: queries where the searcher wants to do something, not to read something.

This is the useful conclusion for a product site. Pages that exist to catch a definition search were always the weakest part of a library; pages that answer a buying question still get the click.

Two moves, in this order. The first makes you the page that gets cited; the second stops you spending the next quarter writing pages nobody will click.

1. 1

   Make the page quotable

   Assistants cite pages that answer the question directly, with clear structure, consistent naming, and lists, comparisons and figures they can quote. Put the answer in the first two sentences under a heading that matches the question.

2. 2

   Let the crawlers in

   The answering crawlers have to be allowed in robots.txt. Blocking them, or blocking Googlebot to avoid overviews, removes you from the results as well as the answer.

3. 3

   Move effort towards queries that convert

   Comparisons, alternatives pages, integration pages, pricing and the jobs your product does. These still send clicks and they send buyers.

4. 4

   Add the thing an overview cannot summarise

   Your own screenshots, your own figures, a tool that runs on the page, a worked example with real numbers. The reason to click through is the part that will not fit in a box.

5. 5

   Track citations by hand

   Once a month, ask the assistants the ten questions your buyers ask, and write down who is named. There is no report for this, so the list is your measurement.

• Do not block Googlebot to stay out of overviews. There is no separate opt-out: the same crawl feeds Search, so blocking it removes the pages from results entirely.
• Do not rewrite a whole library because of one bad month. Check the dates against Google's status dashboard, your own deploys and last year's figures first.
• Do not chase a number nobody publishes. Time spent estimating overview losses is time not spent on the pages that still earn clicks.
• Do not delete informational pages that feed your other pages. They earn internal links, mentions and citations even when they earn fewer clicks.

1. Compare the last 28 days to the 28 before in the Performance report, Queries tab, sorted by click difference.
2. List the queries where clicks fell and impressions held.
3. Search the top five of those in a clean window. Note the overview, its citations and any new rival.
4. Fix one page: the answer earlier, a table added, the heading matched to the question.
5. Ask three assistants your main buying question and write down who is named.
6. Write one line in a log: the date, what you saw, what you changed.

The log is the part people skip and the part that pays. Without last quarter's note, every drop looks new and every recovery looks like luck.