AI Overviews Reach 83% of Branded Searches: What the Data Really Shows

AI Overviews appeared on 82.91% of the branded US desktop results in Ahrefs’ September 29 crawl. The striking part isn’t just the rate. In a separate 100-brand panel, the AI answer often occupied the first result and frequently cited sources outside the brand’s own website.

This is not proof that every navigational query now triggers an AI Overview, or that every brand has lost the first organic click. It is a large, dated observation of one market, one device class, and two related but different samples. Read correctly, it gives brand and search teams a practical new audit target: the answer that appears before the official result.

Three layers of evidence sit behind the 83% headline

Ahrefs published the study on October 2, 2026. Its broad sample contains 232.8 million US desktop search-result crawls from July through September: 92.5 million for branded keywords and 140.2 million for non-branded keywords.

AI Overview prevalence in the Ahrefs crawl set

Month	Branded keywords	Non-branded keywords	Gap
July 2026	61.3%	73.5%	12.2 percentage points
August 2026	65.1%	74.7%	9.6 percentage points
September 2026	73.3%	78.3%	5.0 percentage points

Between July and September, the branded rate increased by 12 percentage points. That is about 19.6% growth relative to July’s 61.3% rate. The non-branded rate increased by 4.8 points, so the gap between the groups narrowed by 7.2 points. Those are calculations from Ahrefs’ published monthly averages, not separate measurements by SearchEngineAnswer.

The study then uses a second panel for position and citation analysis. It checked the exact names of 100 companies selected from Interbrand’s Best Global Brands 2025 list at 14 weekly checkpoints. On September 30, 63 of those 100 exact-name queries returned an AI Overview. Of the 63, 48 appeared in position one (76.2%), with an average position of 1.49.

Keep the two denominators separate. The 82.91% figure comes from the much larger branded-keyword crawl on September 29. The 63% figure comes from 100 bare brand names on September 30. Combining them as one sample would erase a useful distinction between every keyword Ahrefs classified as branded and a small panel of exact company names.

The citation risk is an incomplete third-party narrative

Ahrefs could inspect 57 source lists in the 100-brand panel. Wikipedia appeared in 41, YouTube in 22, LinkedIn in 13, Yahoo Finance in nine, Facebook in six, and Walmart in four. One AI Overview can cite several domains, so these counts do not sum to 57 and should not be treated as shares of all citations.

The most dramatic interpretation would be that competitors are taking control of brand answers. The observed data does not support that conclusion. Only two of 58 fully visible answers explicitly discussed competitors, and Ahrefs verified one competitor-owned website across the 57 source lists. The nearer risk is subtler: an old video, an incomplete directory profile, a marketplace listing, or a stale corporate description can become part of the first answer even when no competitor appears.

That changes the brand-query task. The traditional check asks whether the official site ranks first. The new check asks whether the AI answer is present, where it sits, what it says, which sources it trusts, and whether those sources agree with the current facts.

Translate the citation counts before acting on them

Wikipedia appeared in 41 of the 57 inspectable source lists: 41 ÷ 57 = 71.9%. That describes answer-level source presence. It does not mean that Wikipedia received 71.9% of all citation links. An answer can contain several links to several publishers.

The first-position count has another denominator. The 48 position-one Overviews equal 76.2% of the 63 answers that appeared, but 48% of the full 100-brand panel. For a brand team deciding which searches to inspect first, the latter describes the panel-wide exposure to a top-position answer. Neither calculation predicts clicks for your company. Check the exact query before changing the website.

Run a 30-query brand-answer audit.

Do not start with a visibility score. Build a small panel that reflects the questions a customer, journalist, candidate, investor, or partner might ask before visiting the official site.

1. Five exact-name queries: the brand, common misspelling, parent company, local-market name, and product family.
2. Five trust queries: reviews, complaints, safety, privacy, and customer support.
3. Five decision queries: price, alternatives, comparison, best use case, and who should not buy.
4. Five factual queries: founder, ownership, location, launch date, and current product availability.
5. Five reputation queries: awards, controversies, layoffs, sustainability, and market position.
6. Five action queries: login, contact, cancel, return, and download.

Run the panel in the market and language that matter to the business. Record device, date, signed-in state, exact query, whether an AI Overview appeared, its position, the visible answer, every cited URL, and any factual conflict. Rerunning a week later is more useful than extrapolating from a single screenshot.

Minimum row for a branded AI Overview audit

Field	What to preserve	Why it matters
Query context	Exact query, country, language, device, date, signed-in state	AI surfaces and sources can vary by context.
Surface state	Present or absent, collapsed or expanded, and position	An answer below the first result creates a different risk.
Answer claims	Material facts, recommendations, caveats, and omissions	Visibility without accuracy can be harmful.
Sources	Source URL, publisher, date, ownership, and freshness	The citation list reveals which external records need correction.
Business action	Correct, update, contact publisher, monitor, or no action	The audit should end in a defensible decision.

Repair the source graph before chasing another score.

If an answer is wrong, first find the cited page that carries the wrong fact. Correct owned profiles, help pages, product documentation, executive biographies, location pages, and structured data. When the source is independent, provide a concise correction with a current primary document. Do not ask a publisher to remove accurate criticism merely because an AI system cited it.

Then make the official site easier to verify. Keep names, dates, prices, contact details, policies, and product status consistent across pages. Link the strongest evidence from the pages that answer the corresponding brand questions. This does not guarantee an AI Overview citation, but it reduces contradictions in the public record.

Use the site’s AI visibility measurement crosswalk before buying a monitoring product. A mention, a citation, a position, and a referral visit are different events. If a tool compresses them into one score, keep the underlying rows.

What this study cannot tell us

• It observes US desktop results. Mobile and other countries may differ.
• Ahrefs’ branded-keyword classifier and crawl set are not every Google query.
• The 100-brand panel favors globally recognized companies and should not be treated as a small-business benchmark.
• The article reports prevalence, position, and citations. It does not measure conversion, trust, or the click effect of these specific branded results.
• Google can change the trigger, layout, and cited sources after the captured dates.

The correct conclusion is therefore bounded: branded queries became much more likely to contain an AI Overview in Ahrefs’ US desktop data during the third quarter of 2026. That is enough to justify a brand-answer audit. It is not enough to predict the result for a particular company without checking it.

Sources

• Ahrefs, AI Overviews Currently Appear for 83% of Branded Searches
• Google Search documentation for AI features and website owners