Discussion starter: an AI search question, posted with an answer from Niraj Raut to open the thread. If you have dealt with this on a site, reply with what you saw, especially where it differs.
- AI Search
Zero-click AI answers cut top-of-funnel visits, yet branded search sometimes rises afterwards. Has anyone quantified that effect credibly?
Disclosure: Niraj Raut, who posted this answer, runs the SEO consultancy linked at the end of it.
1 reply
-
adminAdmin
Not credibly, as far as public evidence goes. The loss side has been measured with a transparent method; the branded-search offset has not. Pew Research Center’s browsing data showed that Google users who saw an AI summary clicked a traditional result on 8% of visits, against 15% when no summary appeared. The offset is asserted but not sized: Similarweb’s 2026 article on zero-click marketing, for example, argues that people who don’t click will search the brand later, without putting a number on it. The published claims of a brand lift that I can point to are correlational, and they do not separate AI exposure from everything else that drives brand demand.
The more useful question is whether you can size it for your own brand. Partly. As of September 2026, Search Console’s generative AI performance reports give page-level AI impressions for every eligible site, which, combined with branded query data, makes a first estimate possible. The closest thing to an experiment is an on-off test with Google’s AI features opt-out, and it has a real cost.
What has been measured: the click loss
Pew’s study carries weight because its method is disclosed: browsing data from 900 US adults in its KnowledgePanel Digital panel, covering 68,879 Google searches in March 2025. About 18% of those searches produced an AI summary. Beyond the 8% versus 15% gap, users clicked a link inside the summary on 1% of visits, and ended their browsing session on 26% of pages with a summary against 16% of pages without one.
Its limits matter here. It is US-only, covers one month and Google only, and it reports what happened on the results page and whether the session ended. It does not report whether those users later searched for a brand named in the summary, which is the effect in question.
What has been claimed: the branded offset
Similarweb’s article (published June 2026, updated August 2026) states the hypothesis plainly: many people won’t click at the moment of exposure, but a meaningful share will search the brand name days or weeks later, and that branded search is a measurable downstream signal of zero-click visibility. It offers no figure for the size of the effect and no design that separates cause from correlation.
The mechanism is plausible. Advertisers measure this kind of effect, known as search lift, by withholding ads from a randomly selected holdout group and comparing brand searches between the two groups. Organic AI answers have no holdout, because you don’t control who sees them. That missing holdout is the gap the published claims run into.
Think of an AI answer as a shop window on a street where fewer people now walk in. Some come back later and ask for the shop by name. Counting those requests doesn’t tell you whether the window caused them, because the shop may also be advertising, getting press, or simply growing. You would learn more by covering the window for a few weeks and seeing whether the requests drop. In data terms, the window is AI exposure, the requests are branded searches, and covering the window is the opt-out test described below.
Why the offset is hard to pin down
- Share is not volume. When non-brand clicks fall, branded clicks become a larger share of organic even if branded demand hasn’t moved.
- Reverse causality. If brands that people already search for are more likely to be named in AI answers, rising branded search and rising AI visibility can share a cause, brand growth, without one driving the other.
- Everything else that drives brand demand. Paid media, PR, launches, pricing and seasonality all move branded search, and they keep running while a content program does.
- Measurement breaks. AI Mode activity has been counted in Search Console’s Web totals since June 2025. When Google stopped supporting the &num=100 parameter in September 2025, many sites saw desktop impressions drop, while clicks were affected far less, according to Brodie Clark’s analysis. Year-on-year impression comparisons across those dates need care; clicks and an external index such as Google Trends are safer cross-checks.
- Exposure outside Google. BrightEdge’s August 2026 data attributes about 95.1% of tracked AI referral traffic to ChatGPT (its methodology is not detailed), and ChatGPT gives you no impression data at all. Bing’s AI Performance report, launched in public preview in February 2026, shows citations across Copilot and Bing’s AI summaries, but no clicks.
- Demand that bypasses search. People can type the URL, open an app, or click through from the assistant. GA4’s AI Assistant channel (added in May 2026) groups referred visits from assistants such as ChatGPT, Gemini and Copilot, but visits without a referrer still land in Direct.
Hypothetical example:
A software company had 70,000 non-brand and 30,000 branded organic clicks a month. After AI Overviews expanded on its informational queries, non-brand clicks fell to 50,000 while branded clicks stayed at 30,000. Branded share of organic clicks rose from 30% to 37.5%, and the monthly report called it brand growth. Branded volume did not change at all. Any claim of an offset has to start from absolute branded demand, not its share.
How to estimate it for your own brand
No single method below is proof. Several pointing the same way give you a number you can defend with its uncertainty stated.
Method What it needs What it can show Main weakness Lagged time series Weekly AI impressions, branded clicks and impressions, Google Trends, a log of other brand activity Whether branded demand follows AI exposure with a consistent lag Correlation; shared drivers such as brand growth Topic dose-response AI impressions grouped by topic, and “[brand] + topic” queries Whether exposure on a topic predicts brand searches for that topic specifically Needs enough branded-topic query volume Market comparison Several markets where AI exposure or your content changed at different times A difference-in-differences estimate Markets differ in other ways; small markets are noisy Attribution survey “How did you hear about us?” with AI assistant options, asked consistently Exposure outside Google, including ChatGPT Recall bias and self-report Switchback with the AI features opt-out Willingness to lose Google AI visibility during off periods A near-causal estimate for Google’s AI features Cost, site-wide scope, and it ignores other assistants The set-up for the observational methods:
- AI exposure by week. Export impressions from the Search generative AI performance report. It covers AI Overviews, AI Mode and generative AI features in Discover, with page, country, device and date filters, but no clicks, CTR or queries. It reached all eligible sites worldwide on 31 August 2026. Group pages by topic; the country filter supports the market comparison.
- Branded demand three ways. Search Console’s branded queries filter, checked against your own brand-term regex; Google Trends for the brand name, which does not move with your rankings; and impressions from brand search campaigns if you run them.
- Everything else that moves brand demand. Log campaigns, PR coverage, launches and price changes by week, and exclude or model those weeks.
- Lags, not same-week correlation. Test lags of one to six weeks, in line with the “days or weeks later” pattern the hypothesis describes, and look for a lag that holds across topics and markets. One strong correlation at one lag in one market is noise until it repeats.
The closest thing to an experiment: switching Google’s AI features off and on
Since 31 August 2026, the Search generative AI control in Search Console has been available to websites worldwide. It is a site-level setting that opts the whole site out of AI Overviews, AI Mode and AI Overviews in Discover. Opted-out sites get no traffic or impressions from those features, and Google says the setting will not be used as a ranking signal for search results outside those AI features.
That makes a switchback design possible: alternate off and on periods and compare branded demand between them. Before the first off period, measure how long AI impressions take to disappear from the generative AI report after you switch, and how long they take to return. Make each period at least that lag plus four weeks, so each period yields several weekly readings after the switch settles, and run at least two full cycles so that one campaign or one seasonal week cannot carry the result.
Count the cost first:
During off periods you lose AI impressions and whatever clicks and citations they bring, and competitors may take your place in those answers. ChatGPT exposure continues regardless, so the test measures Google’s AI features only. When the control was first tested with UK publishers, Digiday’s June 2026 reporting described it as leaving publishers with a choice they can’t safely use. For a site that relies on informational content for acquisition, the test may cost more than the answer is worth.
What result should change your content plan
If branded demand falls during off periods and recovers during on periods, by more than its normal week-to-week range, your informational pages are doing brand work and can be valued on branded conversions as well as their own clicks. If branded demand tracks AI impressions in the observational data but does not respond in the switchback, the correlation was probably a shared cause such as brand growth. If nothing moves in either, judge informational pages on their own clicks and conversions, and consolidate the ones where AI answers now meet the need.
Until you have one of those results, keep the offset out of forecasts and board slides. Public evidence does not yet offer a number you could borrow.
Need help with this on your own site?
Niraj Raut works with businesses in Nepal, Australia, the UK, Europe and the US on technical SEO, ecommerce SEO, local SEO and AI search.
Add a reply Cancel reply
Replies are open to approved members. Log in or apply to join.