Ziff Davis earned 2.7 billion AI impressions and lost 37% of its Google referrals

As of November 2026, organic Google referrals to Ziff Davis properties fell about 37% year over year in the third quarter. The same quarter produced about 2.7 billion impressions inside AI Overviews and AI Mode in Search Console. The company's CEO set the two numbers next to each other on an earnings call, and the gap between them is not an AI click-through rate.

If you run an ai seo singapore programme and have just opened the new generative AI report in Google Search Console, this is the decision in front of you: do you treat those impressions as a growth KPI, or as something you read separately from clicks?

The short answer for a marketing lead

Treat the AI impressions figure as a visibility count, not as demand. It sits in a different reporting layer from clicks, and Google publishes no separate count of clicks on AI answers. Anyone who divides one by the other gets a number with no defined meaning.

The practical rule: keep AI impressions, Search clicks, site visits and enquiries as separate rows in your monthly report. Each answers a different question about the funnel, and the ai seo singapore discipline is to report them that way rather than merge them into one growth number.

What Ziff Davis actually reported

Revenue for the quarter fell 5.6% to $289.3 million, a smaller drop than the referral decline. Ziff Davis attributed the gap to businesses that do not depend on Google traffic, such as subscriptions and apps.

On the November 6 earnings call, CEO Vivek Shah said referrals from organic Google search fell about 23% in the same quarter a year earlier. In the quarter just reported the drop was about 37%. He said the company is not planning on a search recovery and expects referrals to keep falling.

Technology & Shopping, the segment closest to search-driven product research, was the one Shah described as facing the strongest search traffic headwinds.

Two earlier data points from the same CEO show how the AI surface has widened. He told analysts that AI Overviews were present in 29% of the queries that generate the majority of Ziff Davis traffic. By August he put AI Overviews at about 50% of the queries relevant to its properties, up from about 36% when he last gave the figure.

Why 2.7 billion divided by clicks is not a click-through rate

Shah said the sites generated about 2.7 billion impressions for AI Overviews and AI Mode in the quarter. He estimated that this was nearly 10 times the clicks those same sites received from Google Search.

The ratio is computable by any site that has the generative AI report. It is also easy to misread, and the definitions explain why.

Google's help page for the report says it covers impressions for AI Overviews and AI Mode. An impression there is a count of how many times links to your site were shown inside those features. The report draws its data from the Web search type in the main Performance report, so AI impressions form a slice of total Search impressions.

Clicks are counted differently. A click on a link inside an AI Overview or inside AI Mode still counts as a click on your site in Search. Total clicks from Search therefore include clicks earned on AI answers plus clicks on every other kind of result. Ziff Davis did not say on the call where its click total came from.

Put together, the division compares AI feature impressions to all Search clicks. The result is an impressions-to-clicks ratio across two different denominators. It can look like a click-through rate without being one, because the numerator counts only AI placements and the denominator counts everything.

For a Singapore SME the arithmetic problem is smaller but the interpretive problem is the same. If your Search Console shows a large AI impressions total and a far smaller Search clicks total in a month, the ratio tells you how many impressions you get per click across all of Search, not how many people clicked an AI answer. Google does not give you that second number.

Three limits built into the report itself

The report has documented limits that matter before anyone builds a target on it.

1. Scope. It counts impressions for AI Overviews and AI Mode. It does not report which questions triggered those impressions, and it does not break out conversions.
2. Availability and data. Google rolled the insights out to all websites worldwide as of August 31, 2026, after announcing them on June 3, 2026. A site without enough AI impressions may not see the report at all, and the newest days can still be preliminary.
3. Row limit. The 1,000 row limitation of the Search performance report also applies here, so a large site that sorts by page loses rows below the cut.

None of these limits makes the report useless. They set what it can answer: how often your links appear inside AI features, on which pages and in which countries. They say nothing about what happened after the impression.

The missing denominator: AI clicks

The UK Competition and Markets Authority has set a Dec. 3 deadline for Google to give publishers impressions, clicks and click-through rate for AI search features in the UK. That deadline is the closest thing to an official AI click count.

Until that data lands, every AI click-through rate you see is reconstructed from a numerator and a denominator that do not belong to the same population. That includes the ratio a large publisher can produce with the same Search Console report.

For a marketing lead, this changes the planning question. You cannot optimise an AI click-through rate you cannot observe. You can ask whether your brand appears in AI answers for the questions that matter, and whether your pages are the ones retrieved and quoted. That is the question a best seo singapore brief should put in writing before any tracking starts.

How to read the AI report without mixing layers

Keep four rows in the same monthly sheet, each sourced from its own place.

The reading rule is that no row is divided by the row above it. The AI impressions row and the Search clicks row move independently; one can rise while the other stays flat. When that pairing shows up, my own reading is that the page is being surfaced more often without being chosen more often — a judgement about the two rows, not a causal rule the report proves. A fall in clicks with stable enquiries means the traffic that disappeared was not converting to begin with. The ai seo singapore page frames the work the same way: find the demand, strengthen the pages, then track what actually moved.

Then take the step no top page does: rebuild the mixer's own arithmetic on one line and write it down, so the number stops looking like a click-through rate. Divide the AI impressions row by the all-Search-clicks row. My own worked example: 2.7 billion AI impressions divided by the clicks Ziff Davis says its sites received is the "nearly 10 times" figure on the call. The numerator counts only placements inside AI Overviews and AI Mode; the denominator counts clicks on every result type in Search, AI answers included. Two different populations, one division. The sentence to write in your report is: "AI Overviews and AI Mode showed our links this month the number in row one, while all of Search sent us the number in row two; these are separate layers and this ratio is not a click-through rate." Every reader who reproduces the division gets the same picture of what the two rows are and are not.

Two habits make the comparison honest across months. Fix the engine, the language, the region and the question set before you compare one period with another. And check where the change sits before crediting any single edit.

If the entry point is a paid campaign rather than organic search, the guide to competing with larger brands on Google Ads covers the paid side of the same funnel.

How Topkee reads an AI impressions number

Topkee's knowledge base treats Google Search Console's generative AI report as a supplement to real impression data rather than a breakdown tool. The report's documented scope is impressions for AI Overviews and AI Mode, and Topkee states plainly that this cannot be expanded into the prompts people typed, the query fan-out inside the engine, or standalone conversions.

That position comes from how the layers are separated in practice. Topkee's performance observation framework splits reporting into four layers: service delivery, search and AI visibility, visits and interaction, and business results. AI impressions sit in the visibility layer, Search clicks sit in visits and interaction, and conversions belong to business results, confirmed against the client's own standard and CRM or analytics data. Being retrieved, being cited, being visited and producing an enquiry are four different stages, reported separately. A single impressions figure cannot stand in for any of the stages above it.

The same boundary appears in Topkee's AI browsing analysis, one of the six research steps of its GEO service. It distinguishes sources the engine retrieved from citations visible in the answer, and it states that neither equals site visits. Real crawler visits, referral traffic and conversion have to come from site data or logs.

For a Singapore business the division of labour is straightforward. Topkee's SEO work counts commercial keywords entering the organic top 10, measured by keyword clusters and target pages. Its GEO work counts whether the brand appears and is described accurately in AI answers, tracked over a fixed question set on sampled engines. The two can run as one plan sharing content. When an AI impressions figure in Search Console looks alarming, the ask is which of those questions changed, not whether the impression number can be converted into clicks.

What to do this month

Start with one page that earns AI impressions and still earns clicks. Compare its AI impressions month over month against its Search clicks from the same report and its sessions in analytics.

If clicks fell and sessions held, the drop is in the click step, not in discovery. If both fell, the page lost ground in Search generally and that is an SEO problem you can fix. If AI impressions rose while everything else was flat, you have evidence the page is being surfaced without being chosen, which is a content and snippet problem rather than a traffic problem.

Run the same comparison against two competitors in your category, using each page's AI impressions and Search clicks from your own account where you have access, and from public data where you do not. A best seo singapore programme that ignores the AI surface will report clicks only; one that tracks the AI surface without clicks reports visibility only. Put the two rows side by side for the page you chose, and read the pair before you read either number alone. When you want the AI-surface question answered with a fixed question set instead of an impressions total, the custom AI agent build page is where that next step starts.

FAQ

Is 2.7 billion AI impressions a lot?

Volume alone says little. For Ziff Davis it was nearly 10 times the clicks its sites received from Google Search in the same quarter. What matters for your own site is the direction of the AI impressions and whether the pages earning them also earn clicks and enquiries.

Can I calculate my own AI click-through rate from Search Console?

Not from the reports Google publishes as of November 2026. The generative AI report gives impressions for AI Overviews and AI Mode, while the Performance report gives total clicks from Search, which include clicks on AI answers and on every other result type. Anyone dividing one by the other is measuring something else.

Why did AI impressions rise while my clicks fell?

More queries can carry an AI answer than did a year earlier. Ziff Davis put AI Overviews at about 50% of the queries relevant to its properties in August 2026, up from about 36% earlier. When a feature that shows links appears on more queries, impressions grow whether or not the reader clicks.

Does the report cover AI Mode and Discover?

The Search version of the report covers AI Overviews and AI Mode. Google says it expects to update that list as Search develops. Discover has its own generative AI report.

Is this a reason to stop investing in SEO?

The referral decline is real for a publisher whose traffic depended on search clicks. A business selling a service has a different funnel: search clicks still reach a page where the reader can enquire. The sensible response is to measure AI visibility separately and keep fixing the pages that still earn clicks and enquiries.

Sources

1. Ziff Davis Says Organic Google Referrals Fell About 37%, Search Engine Journal, 2026-11-09. https://www.searchenginejournal.com/ziff-davis-organic-google-referrals-decline/596279/
2. Ziff Davis Reports Third Quarter 2026 Financial Results, Ziff Davis, Inc. (SEC Exhibit 99.1), 2026-11-05. https://www.sec.gov/Archives/edgar/data/0001084048/000108404826000048/zd20260930pressrelease.htm
3. Generative AI performance report (Search), Google Search Console Help, undated. https://support.google.com/webmasters/answer/16984139
4. Introducing Search Generative AI performance reports in Search Console, Google Search Central Blog, 2026-06-03. https://developers.google.com/search/blog/2026/06/gen-ai-performance-reports