Google's AI answers reducing search traffic by 40%

Hacker News by 19 min read 31x views
Google's AI answers reducing search traffic by 40%

Share Post

Google's AI answers are taking clicks from websites in the hunt results. In a randomized experiment, researchers hid AI Overviews from several US desktop users. Comparing them alongside users who motionless saw Overviews shows that an Overview cuts clicks to exterior websites by concerning 40%. Those Overviews now appear on additional than fractional the US keywords DataForSEO tracks, up from 6% two years ago. The industry's answer, GEO (generative motor optimization), has yet to display a way to compensate for the loss. While AI citations correlate alongside the identical benevolent of nexus overview that goes alongside higher Google rankings, in March AI assistants sent websites fair 0.3% of their traffic.

Search motor optimization (SEO) motionless decides whether you can be found, but anywhere an AI answer appears, a ranking brings 40% small visits than it used to. Generative motor optimization (GEO), the custom of getting your satisfied cited in AI-written answers, looks mostly akin SEO affirmative two additions: being present on the third-party sites AI answers diagram from, and penning pages that are uncomplicated to quote. We established little evidence for any another tactic.

Google answers first

An AI Overview is a generated summary that Google places complete the average results, alongside links to the pages it drew on. In DataForSEO's monthly panel of 20 myriad keywords, the portion of results showing one climbed from 5.93% in September 2024 to 61.43% in August 2026. The diagram at the top of this article shows that climb. Over the identical duration the featured snippet, the quoted passage Google used to display at the top of a page, cut from 8.96% to 0.44%, and the discussions and forums obstacle cut from 21.13% to 7.42%. Video results rosy from 9.67% to 19.98%, although that is downward from a highest of 31.05% in October 2025, and local map listings rosy from 19.98% to 25.99%. DataForSEO's dataset covers 190 myriad keywords, 165 myriad AI answers, and 1.1 milliard backlinks, mostly for US English searches. Pew Research Center's data gives an autonomous check. Pew logged the searches of 900 US adults in March 2025, re-ran them in April, and established summaries on 18% of them. DataForSEO's figures for those months are 14.67% and 17.63%, near to Pew's.

You can't reliably evade Overviews by choosing distinct keywords. Between the fourth fourth of 2025 and the third fourth of 2026, the charge rosy for all category of hunt intent:

Search intentQ4 2025Q3 2026
Informational (looking item up)48.42%70.44%
Commercial (comparing options)26.30%51.35%
Navigational (looking for a site)17.04%34.03%
Transactional (ready to buy)10.82%28.64%

Low-volume keywords aren't safer either. In the third fourth of 2026, keywords alongside 201 to 300 monthly searches showed Overviews 58.59% of the time, against 48.62% for those complete 600. A distinct dataset, from June to September 2026, covers difficulty and price. The easiest keywords showed Overviews 60.64% of the period and the hardest 53.76%. Keywords costing advertisers additional than $5 a click were at 49.55% and those under 25 cents at 50.54%, alongside a low of 42.63% between $0.51 and $1.00. More than fractional the outcome pages have no cost data and display Overviews 60.06% of the time. The charge does change by industry. Health is at 69.68% and backing at 68.80%, during apparel is at 37.48%, dining at 29.37%, and genuine asset at 22.27%.

Queries are additionally getting longer, particularly on desktop. Over a year, queries of five or additional words grew from 29.3% to 34.8% of distinctive desktop queries and from 29.6% to 31.9% on mobile. Three and four term searches are motionless the most common.

What happens to the clicks

The strongest evidence is an experiment. Saharsh Agarwal and Ananya Sen gave US desktop Chrome users an expansion that put all of them at random into a group. One collection saw Google as usual, and another had AI Overviews hidden. The researchers afterward watched two weeks of genuine searches in January and February 2026. On searches that would display an Overview, the collection that could see them averaged 0.37 clicks to exterior websites per search, against 0.62 for the collection that couldn't, which the authors study as a 39.8% reduction. Searches ending alongside no outbound click went from 54% to 73%. Clicks on ads didn't change, which suggests Google takes attention not to cannibalize its own revenue. Across all searches, alongside and without an Overview, the authors put the loss at approximately 18.5% of the integrated clicks a location gets from Google. The authors anticipate the defeat to develop as Overviews appear on more searches. This is a randomized controlled trial, the most dependable benevolent of study for separating cause from association or coincidence. The consequence was statistically important for informational searches. The navigational and transactional samples were too small to resolve anything. This is a operating paper, not yet equal reviewed, and it covers desktop only. A distinct preregistered test by Wang and colleagues, alongside 1,100 people, additionally established that removing AI Overviews and AI Mode raised clicks to publishers.

Observational data points the identical way. Pew established group clicked an average outcome on 8% of visits to a results page alongside a summary and on 15% without one, and ended their browsing meeting on 26% against 16%. Pew's difference isn't a controlled experiment, since the searches that diagram a summary lean to be longer and phrased as questions.

Click-through charge (CTR) is the portion of group shown a outcome who click it. Across more than 2,000 websites in DataForSEO's panel, the desktop CTR of the archetypal integrated outcome cut from 17.83% in September 2025 to 7.43% in August 2026, a autumn of 58%. Mobile cut from 25.17% to 20.80%, a autumn of 17%.

Line diagram of stance 1 click-through charge by month, September 2025 to August 2026. Desktop falls from 17.8% to concerning 15% by February, drops to concerning 8% in March 2026, and ends at 7.4%. Mobile falls from 25.2% to 13% by February, afterward recovers to 22.5% in May and ends at 20.8%.
Desktop misplaced nearly fractional its click-through charge in one month, March 2026, and stayed down. Mobile cut concerning as far complete five months, afterward restored most of it. These figures arrive from Table A2 of the report.

Desktop plunged from 15.18% to 7.94% between February and March 2026. AI Overviews can't effortlessly explain that step since the study has them nearly flat that month, at 42.79% and afterward 43.38%. A alter in the format or norm of the Overviews power explain part of it, although that is speculation on our part. Mobile, meanwhile, sank to 12.96% and afterward rebounded to 22.49% by May. The study itself says its data "does not establish" whether AI answers, layout changes, or changes in behavior are the cause. Its example may additionally be shifting underneath it, since the distinctive desktop queries it counts cut by a third complete the year, from 1.97 myriad to 1.32 million. Whatever the cause, a top ranking on desktop now earns less than fractional the clicks it did a twelvemonth earlier.

DataForSEO additionally estimates anywhere clicks go on pages alongside and without an Overview. By its model, the archetypal integrated result holds 9 to 16 points small click portion on pages alongside one, although these are distinct searches on a sole day's snapshot:

Search intent Position 1, no Overview Position 1, alongside Overview Overview's cited links
Informational37.1%20.7%42.4%
Navigational33.4%24.1%44.8%
Transactional37.6%21.4%43.0%
Commercial33.8%21.1%42.2%

The click portion for the Overview's cited links looks promising, but it comes from a model. Pew, which logged real browsing, saw group click a nexus inner a summary on fair 1% of visits, against 8% for average results. Eighteen months and a ample alter in the product distinct the two measurements, so it is difficult to say which is nearer to the truth.

Google disputes that hunt traffic to websites is declining. In August 2025 its caput of Search, Liz Reid, wrote that total integrated clicks were "relatively stable year-over-year" and that Google was sending "slightly additional norm clicks", definition clicks where group don't quickly arrive back. She gave no figures to rear that up.

Your location can't see whether person goes rear to Google, so you can't inspect Google's quality-clicks assertion yourself, but you can inspect what your Google clicks are worth. RevScope, our product, shows your income per visitor from Google, AI Overviews included, and which landing pages acquire the most.

Who the answers cite

If group peruse the summary and halt there, it matters which sites the summary cites. The study compares Google's AI Overviews alongside ChatGPT. Their most-cited lists overlap, but a distinct benevolent of location leads all one.

RankGoogle AI OverviewsCitationsChatGPTAnswers citing it
1youtube.com51.9Mwikipedia.org4.88M
2reddit.com25.1Mreddit.com3.25M
3facebook.com20.4Mfandom.com670K
4wikipedia.org19.9Mbritannica.com486K
5google.com15.9Mnih.gov476K

The study warns that the two platforms' counts aren't comparable, so difference anywhere a location ranks inside all list. Reddit and Wikipedia position near the top of both, and by our figure 16 of the 30 domains in all top 30 appear on the two lists. The differences are motionless large. YouTube is archetypal in AI Overviews and nineteenth in ChatGPT. Facebook, Instagram, and TikTok create Google's top ten and young female ChatGPT's top 30, during Reuters, The Guardian, AP News, and The Washington Post make ChatGPT's top 30 and young female Google's. ChatGPT additionally frequently cites nobody: 27.3% of its answers had no source, against 0.7% for AI Overviews. So, as the study puts it, power on one phase "doesn't automatically guarantee" the identical on another.

Reddit additionally appears in Google's average top ten for 32.15% of the report's tracked results, nearly eight times Quora's 4.11%. Two limits use to all of this. The archetypal is that the answers arrive from prompts DataForSEO collected, and the report doesn't explain how they were chosen or whether they equivalent what group really ask. The second is that big platforms do concerning as fine in average results: Pew established Wikipedia, YouTube, and Reddit made up 15% of the sources in AI summaries and 17% of those in norm results. Still, the average condemnation that AI favors Reddit holds up in this data.

GEO looks mostly akin SEO

Google's stance is that there are "no additional requirements to appear in AI Overviews or AI Mode, nor another particular optimizations necessary", beyond a leaf being indexed and eligible to display alongside a snippet. Since that is a business describing its own product, we looked at exterior data on two questions. The archetypal is which sites get considered, and the second is which leaf on a location gets picked.

Which sites: nexus profiles correlate alongside citations

A referring domain is another website that links to yours. DataForSEO measured how closely referring domains track AI citations using Spearman's correlation, a mark anywhere 0 method no relationship between two rankings and 1 method they equivalent exactly. The mark was 0.406 for AI Overviews and 0.377 for ChatGPT. Links marked as coming from user-generated content, specified as forum posts, scored lower, at 0.342 and 0.314.

These are average relationships, and they don't display that links logic citations, since large, well-known sites tend to have additional of everything. The samples additionally merge lone domains cited at smallest once, so they can't inform you what it takes to get cited for the archetypal time. A akin form shows up in average rankings: domains ranking archetypal had a median of 111 referring domains against 85 for positions 4 to 10, alongside dense overlap between the groups. For AI Overviews, the median cited domain had 282 referring domains on the easiest keywords and 1,571 on the hardest, following the study excluded elephantine platforms akin YouTube and Amazon and any domain alongside small than 50 referring domains. Small sites weren't measured at all. The ranges are additionally wide, at 518 to 5,765 for the center half of domains on the hardest keywords, so sites alongside extremely distinct nexus profiles get cited.

Which page: cited pages frequently don't position for the query

Ranking for a query and being cited for it are drifting apart. Ahrefs established 76% of pages cited in AI Overviews classified in the top 10 for that query in July 2025, and concerning 38% in March 2026. The two studies differ in sample, parsing, and the Gemini example rearward the Overviews, so the size of that autumn isn't a spotless measurement. One possible scheme is what Google calls query fan-out: its records says these features may matter "multiple connected searches throughout subtopics and data sources". A leaf could afterward be cited since it ranks for a connected inquiry the audience never typed. Ahrefs offers this lone as a possibility, and its before study tested the idea and established that the data didn't assistance it.

The most careful test of page-level strategies we established is a study by the SEO application business SISTRIX, published in October 2026. It paired 48,527 pages cited by ChatGPT, AI Mode, or AI Overviews alongside uncited pages from the identical site, which removes the site's general power from the comparison. In the table, a aspect of 2.7 method pages alongside the characteristic were cited 2.7 times as frequently as pages without it.

Page featureFactorWhat SISTRIX says concerning it
Current twelvemonth in the title2.7A coarse stand-in for maintained content, not a heading trick; an first indication
Table of contents alongside jump links1.7Only 14% of pages had one; 2.5 on editorial sites
FAQ organized data1.5At 833 URLs it pointed the contrary way, a alert concerning small GEO studies
Visible last-updated date1.31.9 on editorial sites, but 0.7 on shops; lone 5% of pages showed one
Ten or additional external links1.3Linking out to sources
Product organized data0.80.5 in AI Overviews, but "not a lever, but no logic for involvement either"; it should stay
Many features combined2.1Against barely optimized pages

An test in the document that gave GEO its name points the identical way as SISTRIX's correlations. Aggarwal and colleagues rewrote web pages and measured how much of a generated answer drew on them. On their 10,000-query benchmark, adding quotations raised a page's visibility by concerning 41%, adding data by concerning 31%, and citing sources by concerning 27%, during keyword stuffing lowered it by concerning 8%, all on their chief measure, the position-weighted portion of the answer's words. Lower-ranked pages gained the most: citing sources lifted a leaf classified fifth by 115% and cut the top-ranked leaf by 30%. They additionally tried Perplexity, an AI hunt product, on 200 samples, uploading the origin content as records fairly than letting it search. The gains on that measure were smaller, at concerning 21% for quotations and 9% for statistics. The chief test used GPT-3.5 study the top five hunt results, which is fine rearward today's products, so the percentages are dated equal if the direction motionless holds.

The most debated shortcut has nothing rearward it so far. llms.txt is a projected document that tells AI systems what a site contains. SE Ranking analyzed nearly 300,000 domains and established no association alongside citations. SISTRIX additionally concluded there is "no measurable effect", although it cautions that domains alongside the document differ from others in many ways, so this part of its data doesn't have the same-site pairing. Google says "you don't need to create new device readable files, AI content files, or markup" for its AI features. The data agrees concerning the files, although SISTRIX's FAQ structured data outcome suggests several markup may help.

AI sends small traffic, but it is expanding and may change well

AI assistants sent 0.28% of total traffic throughout 74,752 websites in March 2026, according to Ahrefs. That was concerning 3.5 myriad visitors, against 345 myriad from Google, or approximately a hundred to one. Ahrefs describes it as small but expanding fast. The true fig is slightly higher since several visits, particularly from mobile apps, get counted as straightforward traffic. The identical Ahrefs part reports that on its own site, AI hunt visitors were 0.5% of visitors and 12.1% of signups. Ahrefs sells SEO software, so its spectators may be unusually apt to use AI tools, and one location proves little. Still, on that location an AI visitant was concerning 24 times as apt to sign up as the average visitor. Some of the value never shows up as traffic. People peruse an AI answer, form an opinion, and never click.

To figure what AI visits are value on your own site, RevScope reports whenever ChatGPT, Claude, Perplexity, and another assistants dispatch visitors and whether they buy. It additionally tracks whenever AI and hunt motor bots crawl your pages.

Google's own reporting has a gap too. It counts traffic from its AI features inner the general "Web" numbers in Search Console. Since 2026 Search Console additionally has generative AI reports, which Google says display how frequently your URLs appeared in those features, alongside alongside pages, countries, and devices. The notice describes impressions and doesn't mention clicks, so you study how frequently you appear, not how many visits or revenue that brings. Rank tracking can't inhabit the gap, since a leaf can clasp stance 1 and acquire small clicks than it used to.

What to do concerning it

  1. Keep the fundamentals. Pages must be indexable and eligible for a snippet. Referring domains correlated alongside citations additional powerfully than user-generated links did, and higher-ranking domains had moderately additional of them, so links from another sites remain a sensible investment, equal although neither association proves cause.
  2. Find out who gets cited on your topics, per platform. Check your crucial queries in AI Overviews and ChatGPT and note the sources. Being described accurately there may matter as much as your own page. Take part in the communities those answers cite, specified as Reddit, so that you are mentioned there.
  3. Write pages value quoting. Answer connected questions directly, add first data and named sources, and use apparent headings. If you run an editorial site, display whenever the leaf was final updated.
  4. Skip the shortcuts. An llms.txt does no damage but has shown no benefit, AI-specific markup isn't needed, and keyword stuffing lowered visibility in the GEO paper's test.
  5. Judge the activity by revenue. With small clicks to go around, it matters additional which ones pay. Tie each conduit to the revenue it produces fairly than to its position or its sojourn count.

The numbers complete are chiefly US English, and the products alter quickly, so the 2025 figures may already be out of date. Nearly all the ample datasets arrive from companies that market SEO data or software, including DataForSEO, Ahrefs, SISTRIX, and SE Ranking, and we market analytics. Almost everything concerning what earns a citation is correlation. Three findings clasp up. Controlled experiments display small clicks to exterior websites whenever Google's AI features are present, domains alongside additional referring domains lean to be cited additional frequently among those already cited, and we established no proven shortcut to being cited.

Sources

Search Insider: 2026 Edition
DataForSEO · Annual investigation study · October 2026
DataForSEO collects hunt and AI answer data for a living and publishes its tables and methods, which allowed us to inspect the monthly figures rearward its headlines. It sells this data, its answer prompts aren't explained, and its click shares are modeled. We depend on it for what appears where, and treat its click estimates and correlations as indications. Our charts are redrawn from its published values.

The Impact of Google AI Overviews on Publisher Traffic and User Experience: Evidence from a Field Experiment
Saharsh Agarwal and Ananya Sen · Working document · SSRN · outline of June 2026
University researchers ran a preregistered randomized test on 1,065 US participants, 770 of them in the two groups compared here. Random project makes it the finest evidence that Overviews themselves decrease clicks. It isn't yet peer reviewed, and it covers desktop Chrome for two weeks. We peruse a copy of the June outline hosted exterior SSRN.

AI in Search Reduces Publisher Referrals Without Improving User Experience: Experimental Evidence
Stephanie T. Wang, Jeffrey Gleason, and colleagues · Preprint · arXiv · August 2026
A second preregistered experiment, alongside 1,100 people, which independently supports the direction of the result. We read its theoretical only, so we use it for that direction and not for any figure.

Google users are small apt to click on links whenever an AI summary appears in the results
Athena Chapekis and Anna Lieb · Short peruse · Pew Research Center · July 2025
A nonprofit investigation institution alongside no hunt merchandise to sell, logging the genuine browsing of 900 consenting US adults. It covers one month, March 2025, re-collected the summaries in April, and doesn't authority for which searches diagram a summary.

AI Features and Your Website and the Search Console reports announcement
Google Search Central · Documentation, final updated December 2025, and blog post, June 2026
The chief sources for what Google says its AI features necessitate and how it reports them. They are dependable on those rules, and they aren't autonomous evidence of what works.

AI in Search is driving additional queries and higher norm clicks
Liz Reid · Company blog · Google · August 2025
Google's community stance on total clicks, from the individual operating Search. Google can see the true numbers, but this post publishes none of them, so we study it as a claim.

Citation Worthiness: Which GEO Tactics Hold Up Across 96,418 URLs
Johannes Beus, Julia Weissbach, and the SISTRIX Content Team · Industry study · SISTRIX · October 2026
Its same-site pairing and correction for multiple evaluation create it additional careful than most GEO studies. It sells SEO software, covers lone German-speaking markets, and reports correlations.

GEO: Generative Engine Optimization
Pranjal Aggarwal, Vishvak Murahari, and colleagues · Conference document · KDD 2024 · type 3, June 2024
The scholarly document that named the site and ran controlled rewrites of pages. It shows satisfied changes can move visibility, but its test motor is now dated, and its authors' "up to 40%" should not be peruse as a current forecast.

Update: 38% of AI Overview Citations Pull From The Top 10 and 76% of AI Overview Citations Pull From the Top 10
Louise Linehan alongside Xibeijia Guan · Industry studies · Ahrefs · March 2026 and July 2025
Ahrefs sells SEO tools and tracks AI answers through its Brand Radar product, current matching concerning 4 myriad cited URLs against rankings in the afterward study. The two studies aren't like-for-like. The 38% counts results in all part of the page; for average links solitary it is 37.1%.

AI Chatbot Traffic: What It Is, and How to Get More
Mateusz Makosiewicz · Company blog · Ahrefs · May 2026
The origin for the 0.28% figure, from analytics on 74,752 sites, and for the signup fig from Ahrefs' own site. It gives conflicting percentages for Google's portion in distinct places, so we difference visitant counts instead.

LLMs.txt: Why Brands Rely On It and Why It Doesn't Work
Yulia Deda · Industry study · SE Ranking · leaf dated October 2026
A ample association study of nearly 300,000 domains. It doesn't name the AI platforms it sampled, so we use it only alongside SISTRIX and Google's statement.

Other Article Hacker News
↑
Close Right Ads
Close Left Ads