AI Search Visibility Report 2026

Search clicks: fewer visits to traditional results alongside AI summaries

Pew analysed 68,879 unique Google searches. Traditional-result clicks followed 8% of visits with an AI summary, versus 15% without one; summary-source clicks occurred on 1% of visits with a summary. The traditional-result gap is 7 percentage points, calculated from the published rounded values.

Source: Pew Research Center — Google users and AI summaries, 22 July 2025

Rankings: first position does not secure the same click yield

Ahrefs reports roughly 58% lower position-one desktop CTR against a modelled counterfactual. Its AI Overview cohort’s observed CTR was about 1.57%, versus an estimated 3.73% without the feature, using the comparison cohort’s trend.

Source: Ahrefs — AI Overviews and desktop CTR, 4 February 2026

Purchases: arriving AI visitors can be commercially valuable

Adobe reports that AI-referred visits to US retailers converted to purchases 42% better than non-AI visits in March 2026. Its broader analytics coverage exceeds one trillion retail visits; that is not the size of the AI-only March subset. The public article does not give the absolute conversion rates or that subset’s count.

Source: Adobe Digital Insights — US retail AI traffic, 16 April 2026

Scale: identifiable AI referrals remain a distinct, smaller stream

Ahrefs’ September 2026 benchmark reports AI assistants at 0.43% of tracked traffic and search at 26.82%. It uses sites participating in its free analytics product, not a representative census of the web. The page was updated on 15 October.

Source: Ahrefs Web Analytics — AI vs Search, updated 15 October 2026

Webflow: AI’s share of signups increased

Webflow’s SEO lead reports AI sources accounting for 8% of new signups in June 2025, compared with 2% in October 2024. The page was updated in February 2026; the outcome dates remain 2024–25. The disclosure supports a growing signup contribution, not a proven increase in paying customers.

Our reading: a share can rise because its numerator grows, its denominator shrinks, or both. Without signup counts and a stable attribution method, do not turn this into a fourfold increase in total business. Ask which landing pages attract qualified users and whether those users activate or pay.

Source: Webflow — Inside its AEO strategy, updated 24 February 2026

Ahrefs: a small traffic source contributed disproportionately to signups

In its June 2025 disclosure, Ahrefs attributes 0.5% of traffic and 12.1% of signups to AI search over the preceding 30 days. The window’s exact endpoints and absolute counts are not stated in the article text. These are shares of two different totals, not a 12.1% visitor conversion rate.

Our reading: this is a reason to inspect the quality of a small channel. It is not a reason to project another site’s signup efficiency onto a new brand. Ahrefs has an established product and publishes the analytics tool used in the case; both context and commercial interest matter.

Source: Ahrefs — Its own AI referral conversions, 16 June 2025

Glasp: the control group grew too

A company-authored preprint compares an optimised /youtube/ corpus with other pages on glasp.co. Reported ChatGPT referrals grew 6.1× for treated pages and 3.5× for controls from January to May 2026; May ends on the 28th. The conservative placebo test gives p=0.16, so the authors describe the intervention evidence as suggestive.

Our reading: raw growth includes a platform tailwind. This case is more useful as a measurement warning than a promise of uplift. Groups were not randomised or matched, several changes were bundled, bot filtering changed, and absolute counts were withheld. The public metadata dates and manuscript display date also differ; we identify the version rather than treating it as a newly verified experiment.

Source: Watanabe & Nakayashiki, Glasp — AEO and platform growth, arXiv preprint v1

Google and Microsoft expose different units

Google’s dedicated report covers impressions in AI Overviews and AI Mode; its launch documentation records worldwide rollout on 31 August 2026. Property and page aggregation differ, and unavailable display values can become zeroes on export. Preserve those distinctions.

Bing’s report counts citations across supported experiences. Citation activity does not state placement or unique audience. Adding it to Google impressions produces a total without a coherent unit.

Source: Google Search Central — Search Generative AI reports, 3 June 2026; rollout note 31 August 2026

Source: Google Search Console Help — Generative AI performance report (Search)

Source: Microsoft Bing — AI Performance public preview, 10 February 2026

Crawl access establishes eligibility, not a result

The reviewed ChatGPT, Perplexity and Claude documents describe source citations and crawler controls. Google also documents eligibility requirements. Access is a prerequisite to investigate; it does not establish a recommendation, accurate brand description or customer outcome. Separate search crawling from model-training controls when checking settings.

Source: Google Search Central — AI features and your website

Source: OpenAI — Searching the web with ChatGPT

Source: Perplexity — What is Pro Search?

Source: Perplexity — Crawlers

Source: Anthropic — Enable and use web search

Source: Anthropic — Web crawlers and website controls

Your analytics configuration changes what you can see

GA’s channel definitions distinguish AI Assistant traffic from Google AI features classified as Organic Search. Filtering only one label therefore misses part of the picture. Reporting identity also affects user counting. Maintain a source/medium view alongside channel groups and document your identity settings; neither reconstructs every unseen answer.

Source: Google Analytics Help — Default channel group

Source: Google Analytics Help — Reporting identity

Define the outcome and freeze the baseline

Select a market, language, landing-page set and outcome. For an exporter, “qualified” might require a target country, product fit and a valid business enquiry. State that rule before seeing the results. Preserve the preceding period, source/medium definitions, consent settings and sales-cycle lag.

Change a bounded set of pages and retain a comparison

Where feasible, randomly hold back comparable eligible pages. Otherwise record why the groups differ and inspect their pre-change trends. Log the exact changes and dates; avoid simultaneous pricing, paid-media and content changes if you want an interpretable comparison. A 60-day operational review is a checkpoint, not a guarantee of statistical power.

Keep a separate answer-observation record

If checking AI answers, freeze the prompt panel, language, market, product/model and search mode; repeat at recorded times. Code brand mention, linked citation, factual accuracy and recommendation separately. A share of this panel is a sample statistic, not market share. Save the answer evidence and failures; do not silently remove unfavourable results.

Judge value only after the outcome matures

Report numerator and denominator, not only a percentage: qualified sessions divided by eligible measured sessions, with both counts. Keep CRM leads deduplicated under a documented rule. Where incrementality is credible, compare incremental contribution after fulfilment costs with content, distribution and measurement spend. Without that evidence, label the result attributed performance rather than ROI.