Most of the Google clicks you lost to AI Overviews probably never brought you a customer. The smaller loss is the one I would investigate first: a buying-intent query where the answer now cites a competitor instead of you.

Here is a typical composite, not one client’s account. A business ranks on page one for 40 or 50 useful questions. Its clicks fall 25 to 30% over a quarter, but impressions hold and positions barely move. The owner sees a red line and assumes the rankings broke. The agency proposes more content to win the traffic back.

That is the wrong call if the disappearing clicks mostly came from definitions, health-style explainers and short how-tos that Google now answers on the results page. Ahrefs’ France comparison of 963 domains found that pattern after the July 2026 launch: highly exposed sites lost 23.1% CTR, and one extreme case gained 183% more impressions while getting 1.9% fewer clicks.

The question is not how to restore every visit. It is which lost visits used to lead to an enquiry, and whether an AI answer now points the buyer somewhere else. Fix that smaller pile first.

The setup: the ranking stayed, but the click moved

Google started answering more questions on the results page instead of sending searchers to the sites beneath it. Pew tracked about 68,000 real searches and found that users clicked a traditional result 15% of the time without an AI summary and 8% of the time with one. About 1% clicked a link inside the summary itself. Ahrefs puts the longer trend at position-one CTR down 34.5% in spring 2025 and 58% by the end of that year.

So a page can keep its place and still lose visits. That is the part that makes the composite business’s report look like a ranking problem when it is not. The searcher sees the answer before reaching the blue links, and fewer people need to click through.

But the average conceals the decision that matters. In the France data, AI Overviews appeared on 22.2% of health-style explainer searches, and CTR for those queries fell 20.4%. A short answer can satisfy that search on the spot. A search for a price, a tool, a booking or a human opinion is different: the answer may inform the choice, but it does not necessarily finish it.

Those two kinds of query do not deserve the same recovery budget. If I blend them into one traffic line, the biggest click losses look like the biggest business losses. They may be nothing of the sort.

The first sign: impressions hold while CTR falls

The owner opens Search Console and sees clicks down month over month. That is a real change, but it is not yet a diagnosis. I would compare the last 3 months with the same 3 months last year and switch on clicks, impressions, CTR and position together. Then I would look at the queries behind the decline.

In this composite, many of the worst click losers still have steady impressions and steady positions. CTR is what fell. That pattern points to fewer searchers choosing the result, not to the page sliding down the rankings. A query that also lost position belongs in a separate pile. It may need a ranking fix, and calling every loss an AI Overview problem would be just as careless as calling every loss a ranking problem.

The Generative AI report does not settle it on its own. Google made the report available to all properties on August 31, 2026, after a June announcement, but it shows impressions only by page, country and device. It has no clicks, CTR, position or query dimension. Clicks from AI features remain pooled in the overall Web totals.

That leaves an unglamorous bit of work: line up the pages receiving AI impressions with the queries and pages losing clicks. The report can tell you where to look. It cannot tell you which lost click once became a customer.

The wrong call: buy back every lost visit

The agency’s proposed content blitz has a tidy logic. Clicks fell; publish and rewrite more pages; chart the returning sessions. I used to reach for more content when traffic dropped, too. For this kind of drop, that reflex spends money before it identifies the loss.

Start with the 40 or 50 question pages in the composite. If Google now answers many of those questions directly, polishing every explainer may make the work look busy without bringing back the people who request quotes, book calls or check out. The business could recover some visits and still leave the buying queries untouched. The report would improve while the problem remained.

That is why the first rewrite should not go to whichever page lost the most clicks. It should go to a page connected to enquiries, if the answer now sends the buyer toward a competitor or a source that excludes the business. The ordering sounds obvious once you say it aloud. It is less obvious inside a dashboard sorted by traffic decline.

The damage: a busy report, an untouched buying query

The broader traffic losses make the wrong call easier to sell. Reddit reported in its Q3 2026 earnings that search referrals fell by more than 1 million per week on average in October, alongside a decline in US daily actives from 53.2 million to 50.7 million. Publishers also saw Google referrals fall, with Chartbeat tracking a 33% global drop from November 2024 to November 2025.

Those figures show how large the shift can look from a distance. They do not tell the owner of our composite business which page to fix on Monday. That requires separating lost attention from lost demand.

The damage in this postmortem is not simply the missing traffic. It is the decision to spend on the most visible decline while a smaller buying-intent query goes unchecked. If the AI answer cites a competitor for that query, the business has a more specific problem than a shrinking sessions chart. The buyer is being shown someone else at the moment they are choosing.

The diagnosis: narrow the loss before you rewrite

I would run one pass through Search Console and the CRM before commissioning a page. It is a short checklist, not an AI visibility programme: reduce a long list of lost-click queries to the ones tied to enquiries, then inspect those answers by hand. The order matters because each step prevents a different kind of wasted work.

  1. Compare the last 3 months with the same 3 months last year. Turn on clicks, impressions, CTR and position so a seasonal dip does not masquerade as a new search problem.
  2. Sort Queries by click difference and keep the AI-shaped losses: impressions steady or up, position steady or better, CTR down. That pattern is the starting filter, not proof that every query has an Overview.
  3. Set aside queries that lost position. Give ranking losses their own job; do not charge an AI-answer rewrite for fixing them.
  4. Tag the remaining queries as informational or buying intent. Comparison, best, cost, near me and product-how-to language belongs in the buying pile.
  5. Match the buying pile to enquiries and closed sales in your CRM or GA4. Use the last-click page path and a 90-day lookback; a query with no connection to a quote, call or checkout moves down the list.
  6. Search each surviving buying query in a clean browser. Record whether the Overview cites you, cites a competitor or does not appear.
  7. Check whether the answer cites a third-party listicle, review site or forum instead of any vendor. If you skip the source check, you can pay for another rewrite of your own page while the answer keeps quoting a page that does not mention you.

The result is a smaller queue, not a promise to recover every lost click. Buying queries that cite a competitor come first. Queries that cite you need watching and, where useful, a focused page improvement. If there is no Overview and rankings held, the AI-answer explanation does not fit that search; investigate it separately.

Paper shop beneath a large glowing search answer box

Even the first pile needs a realistic target. Getting cited can soften a CTR loss without restoring the old traffic. Seer Interactive tracked 3,119 informational queries and found 0.70% organic CTR when a brand was cited in the Overview, against 0.52% when the Overview appeared without citing it and 1.45% when there was no Overview. That is roughly a third more CTR for the cited brand than the uncited one, but still about half the CTR seen without an answer box.

Those are informational queries, not a forecast for the composite business’s buying pages. They are a useful warning against promising a full traffic rebound. Judge the work by citations and enquiries together, not sessions alone.

The fix: work on the source the answer can use

I would start with buying-intent pages that already rank and have lost clicks. That is not a shortcut around the hard work; it is where ranking and citation have the clearest overlap. In an analysis of 1.9 million citations from 1 million AI Overviews, 76.1% of cited pages ranked in the top 10. In one case study, 21 of 22 Overviews cited a page already in the top 10. If a page sits at position 14 for a money query, I would address its ranking gap before betting on a sentence tweak to earn a citation.

For a page that does rank, the work is less exotic than the sales decks suggest. Google’s guide says a page must be indexed and eligible for a snippet to appear in AI Overviews. It does not require special markup, an LLMs.txt file, chunking tricks or a rewrite performed solely for AI. I would make the useful answer easy to find: open a section with a direct response, use headings that match buyers’ questions, put an existing quotable number or table near the top, and connect related pages with links. The point is to make the page clear enough to extract, not to decorate it with new jargon.

But sometimes the answer does not cite a vendor at all. It cites a comparison article, a review site or a forum thread. In that case, polishing the service page alone misses the source Google is using. The work shifts to the cited places: claim the relevant profiles, pitch an update to a list that omits the business, or contribute a useful answer where the discussion is happening. Fix the source of the exclusion, not just the page you own.

I would also check for a block before paying for more writing. A nosnippet tag blocks ordinary snippets and AI use of the page; max-snippet caps snippet length. Pages blocked from the relevant crawlers cannot be fetched in the first place. Keep URLs crawlable and allow OAI-SearchBot separately from GPTBot for ChatGPT search visibility: one governs live search access, the other training. These checks will not make a weak page worth citing, but a blocked page cannot benefit from a stronger paragraph.

What a useful month of help looks like

If the owner hires help, I would expect the first month to produce a short, inspectable queue: which buying queries lost clicks, which ones show an Overview, who is cited, and which lost pages connect to enquiries. Then the operator can work through the ranking gaps, eligible pages, snippet settings and third-party sources that the answers actually use.

An agency that opens with twenty new informational posts and a visibility chart is repeating the wrong call from this postmortem. That plan may rebuild sessions. It has not yet shown how it will recover customers. groas approaches SEO & AI Search as done-for-you execution with a named strategist and logged changes, rather than a quarterly review tied to a percentage of spend. The distinction I care about is whether the work starts with the buying pile and stays accountable to it.

Check those buying queries by hand every two weeks. Note whether the business is cited and whether quote or checkout starts from the relevant pages move over 60 to 90 days. Do not promise the owner the old traffic line. If sessions rise but enquiries do not, the team may have repaired the wrong part of the chart.

The rule I apply now: no page gets rewritten for AI search until I can name the enquiry it once helped produce and check who the answer cites in its place. Everything else waits.