Zero-click, AI referrals or cited clicks: which AI search number belongs in a board report?

Five numbers that count five different things

The AI search figures that reach board packs this year count five different events. They are searches that end without a click, clicks lost when an AI Overview appears, clicks gained when your page is cited, visits referred by an AI assistant, and visits that follow an AI answer by another route. Each is real. None measures the same thing as another, and none on its own answers the question a board asks, which is whether AI answers are helping or hurting the business.

The confusion is understandable. The figures arrive in the same month, from vendors with good data, and the headlines compress them into one story about clicks disappearing. A finance director who reads "zero-click" in one deck and "AI traffic up sixteen-fold"⁴ in the next is right to ask which number to believe.

Our view: both can be true at once, because they measure different things. The job of a board report is to say which event each line counts and what it cannot prove. This piece separates the five measures, then proposes a four-line set that a marketing lead can defend and a finance lead can audit.

Prior work, and our interest

Others have done much of the groundwork. SparkToro has tracked zero-click searches for years, Ahrefs and Seer Interactive have published repeated studies on AI Overviews and click-through rates, and SE Ranking publishes AI referral benchmarks. Our own guide to proving that AI visibility reaches pipeline covers the analytics and CRM fields in detail. This post adds a narrower thing: a map of what each published number counts, and a reporting format for the board.

A disclosure. Sigzen AI sells AI visibility audits and programmes, and our programme dashboard reports several of the lines below. Weigh our view with that in mind.

What does zero-click share count?

Zero-click share counts Google searches that end without a click to any website. It is a measure of how search behaves overall, not of AI answers and not of your brand. In SparkToro's analysis of Similarweb's US desktop and mobile clickstream data for January–April 2026, 68.01% of Google searches ended without a click. SparkToro's 2024 figure was 60.45%, from a different panel, so the comparison is approximate.¹

That is a striking trend, and it belongs in the context paragraph of a board paper. It does not belong in the metrics table. The denominator is every Google search in the US panel, including navigational searches, weather, sports scores and questions answered by a calculator. AI Overviews are one cause among several, and the figure cannot separate them.

Nor can it tell you anything about your own site. A company whose buyers search for specific part numbers lives in a different world from a publisher whose readers ask general questions. Our view: quote zero-click share once, as the reason the board is reading this section, and then move on to numbers you can act on.

How much do AI Overviews cut clicks?

They cut clicks to the top result sharply on the queries where they appear. In Ahrefs' study of 300,000 keywords, comparing desktop click-through rates in December 2023 and December 2025, the top-ranking page's click-through rate was 58% lower when an AI Overview appeared.² Ahrefs sells SEO and AI visibility tools, and the study does not state a geography.

This is the number most often misused in budget conversations. It describes position one on queries that show an AI Overview, averaged across a large keyword set. It is not your loss, because your queries may trigger AI Overviews more or less often than the sample, and your pages may sit lower on the page, where the effect differs.

It is useful for one thing: sizing the risk on your own query set. If you know which of your high-value queries now show an AI Overview, the Ahrefs figure gives a reasonable prior for what happened to their clicks. Then check it against your own Search Console data rather than assuming it.

AI search measures, 2026

Four numbers, four different events

68.01%of US Google searches ended without a clickSparkToro, Jan–Apr 2026 ¹

58%lower CTR for the top result when an AI Overview showsAhrefs, Dec 2025 ²

120%more organic clicks per impression when cited in an AI OverviewSeer Interactive, Jan 2025–Feb 2026 ³

0.32%of visits sent by AI search across 101,574 sitesSE Ranking, Jan–Apr 2026 ⁴

Being cited changes the arithmetic

Being cited inside an AI Overview recovers part of the loss, which makes citation the lever a company can actually pull. Seer Interactive, a search agency, tracked 53 brands from January 2025 to February 2026. Pages cited in an AI Overview earned about 120% more organic clicks per impression than when they were not cited.³

The same panel showed organic click-through rate on AI Overview queries rising from 1.3% in December 2025 to 2.4% in February 2026.³ Seer also found that 95.4% of "X vs Y" comparison queries in its panel showed an AI Overview.³ Comparison queries are where buyers choose between vendors, so for most B2B and retail brands they are the queries that matter.

Two cautions. The panel is 53 brands Seer tracks, not a market sample, so treat it as a strong signal rather than a market average. And a cited page still competes with an answer that may satisfy the searcher without a click. Our view: cited clicks are the best Google-side number to report, because they move when your work moves, but they measure the click you kept, not the influence the answer had.

Is AI referral traffic the number to watch?

It is the easiest to see and the easiest to misread. Across 101,574 websites in 250 countries, SE Ranking found that AI search sent 0.32% of all visits in January–April 2026, up from 0.02% in 2024.⁴ SE Ranking sells search tools, so read this as a vendor study with a large sample.

In the same months, ChatGPT sent 74.78% of those AI referrals and Gemini 11.56%.⁵ So the AI referral line in most analytics accounts is largely a ChatGPT line, and a small one.

The trap is reading a small referral line as small influence. A referral is recorded only when a person clicks a link inside the answer and the referrer survives the trip. Someone who reads a recommendation, closes the app and searches for the brand later leaves no trace of the assistant. Our view: report AI referrals as a floor, label them as such, and never present the growth rate on its own, because growth from a tiny base flatters any trend.

AI-influenced visits: the path referrals miss

AI-influenced visits are the visits that follow an AI answer by another route, and the first panel study suggests they outnumber referrals by a wide margin. Similarweb, which sells traffic and AI visibility data, followed US desktop users in its panel from July to December 2025 across three brand pairs in finance, travel and beauty.⁶ Brands recommended by ChatGPT were 2.5 times more likely to be visited within seven days than brands that were not.⁶

The route matters more than the multiplier. In the same study, 55.9% of AI-influenced visits arrived through search, against 40.4% of other visits, and only 8.8% arrived as an AI referral.⁶ Most of the effect shows up as search visits, most likely searches for the brand, not in the AI channel.

The study is narrow: one engine, desktop only, six consumer brands and one country. It still answers a question boards ask, which is where the AI effect would show up if it existed. The answer is mostly in branded search, which is why that line is in our set below.

How one AI recommendation becomes a visit

1. Buyer asks an AI engine
2. Answer names a brand
3. Buyer searches the name later
4. Visit lands as branded search
5. Form, call or purchase

What each number can and cannot prove

Put side by side, each measure has one honest use and one common misuse. Use the table when someone forwards a headline and asks whether it applies to your company.

Our summary of the five measures discussed above. Sources are footnoted in each section.

Why do the published numbers disagree?

They disagree because they use different panels, periods and denominators, not because one is wrong. A clickstream panel, a keyword set and a brand panel will never produce the same number, and a board paper should say which kind each figure comes from.

Older studies make the point. Pew Research Center tracked the browsing of 900 US adults in March 2025, data now more than a year old. Users clicked a traditional search result on 8% of visits to a page with an AI summary, against 15% without one, and clicked a link inside the summary on 1% of visits.⁷ That is a measure of people, not keywords, and it found a smaller gap than the Ahrefs keyword study above.

Neither figure is wrong. Pew counted what a sample of people did across all their searches; Ahrefs counted click-through rates on the top result across a fixed keyword set. A board does not need to settle that dispute. It needs to know which kind of number it is looking at, and that the company's own data outranks any external figure once there is enough of it.

Our view: when a deck quotes an external number, put the sample, period and geography next to it in the same font size. Most of the arguments we see about AI search impact end once those three facts are visible.

The four lines we would put in front of a board

A board needs four lines: share of answer, clicks on tracked AI Overview queries, branded search demand, and pipeline with an AI touch. Together they cover presence in the answer, the click you keep, the demand the answer creates and the money that follows. Each line carries its limit in writing.

Line 1: share of answer

Share of answer is the proportion of AI answers that name your brand across a fixed set of buying prompts. It is the only line that measures the answer itself. Report it by engine, with a margin of error, because answers vary from run to run. Our measurement spec for AI visibility scores sets out the runs and intervals to ask for.

Its limit: it measures presence on your prompt list, not the market. If the prompts are wrong, the number is precise and irrelevant. Refresh the list each quarter, and show the board three example prompts so it knows what is being counted.

Line 2: clicks on tracked AI Overview queries

Pick the 50 to 200 queries that matter most, note which show an AI Overview and whether you are cited, and track clicks per impression for that set in Search Console. This is the company's own version of the Ahrefs and Seer figures, and it beats both once you have a quarter of data.

Its limit: Search Console does not report AI Overview clicks separately, so the citation status comes from your own checks. Say so in the footnote.

Line 3: branded search demand

Branded search clicks and impressions, weekly, from Search Console's branded query filter.⁸ Given the Similarweb finding, this is where an AI recommendation most often surfaces. Track it by country if you sell in several.

Its limit: branded search also moves with advertising, press and events. Log those dates next to the series, and read the line against share of answer rather than on its own.

Line 4: pipeline with an AI touch

Count opportunities where the buyer named an AI assistant in a "how did you hear about us" field, or arrived through the AI channel, and report their number and value each quarter. The guide to connecting these signals in the CRM covers the fields.

Its limit: self-reported data relies on memory, and most AI-influenced buyers will not say so. It is a lower bound, and the board should hear that in those words.

The four-line board set

1. Share of answer

   By engine, with a margin of error. Limit: only as good as the prompt list.

2. Clicks on tracked AI Overview queries

   Your own clicks per impression. Limit: citation status comes from your own checks.

3. Branded search demand

   Weekly, by country. Limit: also moves with ads, press and events.

4. Pipeline with an AI touch

   Count and value per quarter. Limit: a lower bound, based on memory.

What does a one-page board slide look like?

It looks like four rows, each with a number, a direction, a confidence note and one sentence of interpretation. The example below shows the shape. The company and every number in it are invented.

Illustrative. Fictional company; invented, rounded numbers.

The slide makes no return claim. It says presence rose with confidence, branded demand rose with one confounder, and pipeline with an AI touch is growing from a small base. Our view: that is a result a board can trust, and it sets up a revenue conversation for the following quarter rather than inflating this one.

Questions a board will ask

Boards ask three questions, and each has an honest answer that fits in a sentence. Prepare them before the meeting rather than improvising.

• Are we losing traffic to AI? Answer from Line 2, on your own queries, and quote the external click-loss figure only as context.
• Are competitors ahead of us? Answer from Line 1, by engine, with the competitor set named and the margin of error shown.
• Is any of this worth the money? Answer from Lines 3 and 4 against a prediction written at the start of the quarter, and say plainly when it is too early to tell.

What the board should not hear is a single blended "AI impact" figure. Adding referral visits, estimated influenced visits and recovered clicks into one total double-counts the same buyers and hides which line moved.

What to do before the next board meeting

Start with the lines you can populate from data you already hold, and be explicit about the ones that are still being built.

1. Write one paragraph of context with the external figures, each with its sample, period and geography.
2. Pull branded search clicks from Search Console for the last two quarters, by country.
3. Choose your tracked query set and record which queries show an AI Overview today.
4. Add an AI assistant option and a free-text field to your "how did you hear about us" question.
5. Run a share-of-answer baseline on a fixed prompt list, with repeated runs on each engine.

On our published pricing, the AI Visibility Audit is $3,500 one-time and produces the first line: 150 buyer prompts on ChatGPT, Gemini, Perplexity, Claude, Copilot and Google AI, run through our AI answer monitor on three separate days. The other three lines come from your own Search Console and CRM, and you can start them this week.