What five AI surfaces cite for 40 US questions, matched against Google's organic top 10 on the same day
Opttab Research. Data collected 9 November 2026.
A page-one Google ranking still earns a citation in most AI answers. The odds depend on which AI surface your buyer asks. On 9 November 2026 we asked 40 US shopper and buyer questions in English. Each went to five AI surfaces: ChatGPT, Gemini, Google AI Mode, Google AI Overviews and Bing AI Mode.
The same day, we recorded Google's US organic top 10 for every question. Then we matched every URL each answer cited against that list. We read 196 answers, and 181 of them cited at least one source. Perplexity was planned, but it could not be collected in logged-out sessions that day, so it is not in the results.
This study asks a different question from our Black Friday 2026 study (data collected 31 October 2026) and our Cyber Monday 2026 study (data collected 4 November 2026). Those looked at what AI tells shoppers in six countries. This one looks at where AI answers come from, and how much of that overlaps with classic SEO. Prompts, surfaces and bases differ, so treat every comparison with the earlier studies as directional.
The short version: ranking in Google still gets you into most AI answers. 68.0% of answers with sources cite at least one URL from Google's US top 10 for the same query (95% CI 60.8–74.3%). The best-cited URL sits in positions 1 to 3 in 54.6% of answers. The share runs from 100.0% in Google AI Overviews to 46.2% in ChatGPT. Only 7.1% of answers cite top-10 pages alone. Publishers, review sites and vendor pages fill most of the gaps. Treat SEO as the entry ticket, then earn the rest of each answer surface by surface.
Key findings
Ranking still gets you cited. 68.0% of AI answers with sources cite a URL from Google's US top 10 for the same query. Across all answers, the figure is 62.8%. Matching on domain lifts it only to 75.7%.
Surfaces split sharply. 100.0% of Google AI Overview answers with sources cite a top-10 URL. Google AI Mode does in 57.9%, and ChatGPT in 46.2%.
Ranking rarely owns the answer. Only 7.1% of answers cite top-10 URLs exclusively. Another 25.5% cite sources, but none from the top 10.
Publishers fill the gaps. A publisher or review site is the first non-ranking source in 32.1% of answers. A forum comes first in 1.5%, even though youtube.com and reddit.com are the two most-cited domains.
Freshness overlaps most. 95.5% of freshness answers cite a top-10 URL. Troubleshooting and trust and safety answers do in 60.0% each (indicative).
Best-of answers look current. 40.8% of all answers show a cited source dated 2026. For best-of recommendation questions, the share rises to 79.2%.
What it means: keep investing in rankings, because they get you into most AI answers. Measure each AI surface on its own. Then work the publisher, review and community pages that share the answer with you. To see where you stand today, run Opttab's free AI Visibility Report (https://opttab.com/ai-visibility-report/).
The numbers at a glance
Number | What it measures |
|---|---|
68.0% | of AI answers with sources cite at least one URL from Google's US top 10 for the same query (95% CI 60.8–74.3%, n=181) |
100.0% | of Google AI Overview answers with sources cite a top-10 URL (n=36) |
46.2% | of ChatGPT answers with sources cite a top-10 URL (n=39) |
54.6% | of answers have their best-cited URL in Google positions 1 to 3 |
0 of 196 | answers have their best-cited URL in Google positions 11 to 20 |
7.1% | of answers cite only URLs from Google's top 10 |
32.1% | of answers have a publisher or review site as their first non-ranking source |
18.9% | of answers cite youtube.com, the most-cited domain (reddit.com: 14.3%) |
79.2% | of best-of recommendation answers show a cited source dated 2026 |
Finding 1: Ranking in Google's top 10 still gets you into most AI answers
Across 181 answers that cited at least one source, 68.0% cited a URL from Google's US organic top 10 for the same query on the same day. We computed that figure exactly, with a 95% CI of 60.8–74.3%. Counting all 196 answers, including those with no sources, the share is 62.8%. The lower bound sits above 50%, so we reject our hypothesis that fewer than half of answers would overlap.
We expected domain matching to beat URL matching by at least 15 points. It lifts the share only to 75.7% of answers with sources. When AI cites a site that ranks, it usually cites the page that ranks.
Position matters inside the top 10. In 54.6% of answers, the best-cited URL ranked in positions 1 to 3 (cluster 95% CI 46.7–62.5%). In 16.8%, it ranked 4 to 10. No answer's best-cited URL sat in positions 11 to 20 (0 of 196). Meanwhile, 20.9% cited only pages outside the top 20.
Measure | Answers with sources (n=181) | All answers (n=196) |
|---|---|---|
Cites a top-10 URL (computed exactly) | 68.0% (60.8–74.3%) | 62.8% (55.8–69.2%) |
Cites a top-10 domain (computed exactly) | 75.7% (69.0–81.4%) | 69.9% (63.1–75.9%) |
Cites a top-10 URL (coded field) | n/a | 69.9% (cluster 62.2–77.7%) |
Cites a top-10 domain (coded field) | n/a | 72.5% (cluster 64.6–80.0%) |
We lead with the exact match. The coded top-10 URL field gives a similar 69.9% of all answers. Coders agreed on that field only 71.7% of the time.
Why it matters: Treat SEO as the entry ticket to AEO. A top-three ranking is the most common route into an AI answer, and positions 11 to 20 earned no best-cited spots in this sample. Move the pages sitting just off page one first. GEO / AEO actions in Opttab (https://opttab.com/geo-aeo/) maps each prompt to your pages and lists the content and technical fixes.
Finding 2: Google AI Overviews mirrors the top 10; ChatGPT and Google AI Mode do not
Every Google AI Overview with sources cited at least one URL from Google's top 10. That is 100.0% of 36 answers (95% CI 90.4–100.0%). In 91.9% of Google AI Overviews, the best-cited URL ranked in positions 1 to 3.
ChatGPT sits at the other end. 46.2% of its answers with sources cite a top-10 URL (95% CI 31.6–61.4%). At domain level, that rises to 64.1%. The ChatGPT and Google AI Overviews intervals do not overlap.
We expected Google's own surfaces to lead by at least 15 points. Google AI Overviews does. Google AI Mode, at 57.9%, sits below Bing AI Mode at 75.0%. The intervals for the middle three surfaces overlap, so read their order as indicative.
AI surface | Answers with sources | Cites top-10 URL (exact, 95% CI) | Cites top-10 domain (exact) | Best-cited URL in positions 1–3 (coded, all answers) |
|---|---|---|---|---|
Google AI Overviews | 36 | 100.0% (90.4–100.0%) | 100.0% | 91.9% |
Bing AI Mode | 40 | 75.0% (59.8–85.8%) | 75.0% | 50.0% |
Gemini | 28 | 60.7% (42.4–76.4%) | 64.3% | 27.5% |
Google AI Mode | 38 | 57.9% (42.2–72.2%) | 73.7% | 55.0% |
ChatGPT | 39 | 46.2% (31.6–61.4%) | 64.1% | 51.3% |
Citation volume differs too. Only 28 of Gemini's 40 answers cite a source. By contrast, 91.9% of Google AI Overviews list six or more URLs.
Google AI Mode often cites the right site but a different page. Its overlap jumps from 57.9% at URL level to 73.7% at domain level. That is the biggest gap of the five surfaces.
Why it matters: A top-10 ranking almost guarantees a place in Google AI Overviews. It gives no such guarantee in ChatGPT or Google AI Mode. Track each surface separately instead of trusting one visibility number. Opttab's AI visibility tracking (https://opttab.com/ai-visibility/) reports visibility and cited pages per AI model.
Finding 3: A top-10 page gets you cited, but rarely owns the answer
Being cited and owning the answer are different results. Only 7.1% of answers cite top-10 URLs exclusively (cluster 95% CI 4.0–10.5%, n=14/196). 24.0% draw half or more of their citations from the top 10. Another 36.7% draw some, but fewer than half.
25.5% of answers cite sources, none of them in the top 10 (cluster 95% CI 17.6–33.8%). Another 6.6% cite nothing at all. Many answers cite several pages at once. 36.2% cite three to five URLs, and 30.1% cite six or more.
AI surface | No citations | Cites, none in top 10 | Fewer than half in top 10 | Half or more in top 10 | All in top 10 |
|---|---|---|---|---|---|
All answers (n=196) | 6.6% | 25.5% | 36.7% | 24.0% | 7.1% |
Google AI Overviews (n=37) | 2.7% | 0.0% | 29.7% | 67.6% | 0.0% |
Google AI Mode (n=40) | 2.5% | 27.5% | 20.0% | 27.5% | 22.5% |
Bing AI Mode (n=40) | 0.0% | 32.5% | 55.0% | 10.0% | 2.5% |
ChatGPT (n=39) | 0.0% | 38.5% | 51.3% | 5.1% | 5.1% |
Gemini (n=40) | 27.5% | 27.5% | 27.5% | 12.5% | 5.0% |
The surfaces build different mixes. Google AI Overviews takes half or more of its citations from the top 10 in 67.6% of answers. It never relies on the top 10 alone. ChatGPT takes fewer than half in 51.3% of answers and none in 38.5%. Google AI Mode is the surface most likely to cite only top-10 pages (22.5%, indicative).
Why it matters: Your ranking page will share the answer with pages you don't control. What those pages say about your category and products shapes the answer as much as your own page does. Check which outside pages sit next to yours on each prompt. GEO / AEO actions (https://opttab.com/geo-aeo/) compares your pages with the most-cited ones and flags the content gaps.
Finding 4: Outside the top 10, AI reaches for publishers and vendors, not forums
When an answer cites a page outside Google's top 10, the first such source is most often a publisher or review site. That happens in 32.1% of answers (cluster 95% CI 25.9–38.7%). Brand or vendor sites come next, at 24.0%. Video and social posts account for 10.2%, and forums for 1.5%. We expected forums and video together to outnumber brand and vendor pages. They don't.
Community content still shows up across the citations. youtube.com is the most-cited domain, in 18.9% of answers. reddit.com is second, at 14.3%. 21.4% of answers cite at least one forum or community thread, rising to 64.9% in Google AI Overviews. One reading: the threads and videos AI cites often already rank, so they rarely appear as the non-ranking source.
Coders agreed on the non-ranking source type only 58.3% of the time (kappa 0.458). Treat this split as indicative.
Domain | Answers citing | Share of answers |
|---|---|---|
youtube.com | 37 | 18.9% |
reddit.com | 28 | 14.3% |
support.google.com | 12 | 6.1% |
Software vendor C | 11 | 5.6% |
shopify.com | 11 | 5.6% |
Software vendor B | 8 | 4.1% |
Software vendor A | 8 | 4.1% |
rtings.com | 8 | 4.1% |
consumerreports.org | 7 | 3.6% |
consumer.ftc.gov | 7 | 3.6% |
Each surface reaches for something different. On Bing AI Mode, the first non-ranking source is a publisher or review site in 67.5% of answers. On Google AI Overviews, it is video or social in 43.2%. On ChatGPT, it is a reference, government or education site in 30.8% (indicative).
Across all citations, publishers lead. Of 1,204 answer-domain citation pairs, publishers and media take 30.2%. Software vendors take 22.8%, and review and comparison sites take 14.7%. Forums take 3.6%. In our Black Friday 2026 study, reddit.com was also the most-cited US site, at 13.4% of citing answers. Prompts and bases differ, so that comparison is directional.
Why it matters: Beyond your own rankings, the publishers, review sites and vendor pages in your category are the likeliest extra sources. Make sure they carry accurate facts about you. Answer Engine Analytics in Opttab shows which of those pages AI cites for each prompt, so you know whom to brief first.
Finding 5: Freshness questions overlap most with the top 10; trust and troubleshooting least
We expected explainer and how-to questions to overlap most with Google, and best-of and freshness questions least. Freshness broke that pattern. 95.5% of freshness answers cite a top-10 URL (n=22), the highest of eight intents. How-to follows at 76.0%.
Troubleshooting and trust and safety answers overlap least, at 60.0% each. Best-of answers are where AI most often looks past page one. In 33.3% of them, the best-cited URL ranked outside the top 20. For freshness answers, that figure is 4.5%.
These are coded figures with 22 to 25 answers per intent. Read the order of intents as indicative.
Intent | Answers | Cites top-10 URL | Cites top-10 domain | Best-cited URL outside top 20 |
|---|---|---|---|---|
Freshness | 22 | 95.5% | 95.5% | 4.5% |
How-to | 25 | 76.0% | 84.0% | 8.0% |
Explainer | 25 | 68.0% | 68.0% | 24.0% |
Comparison | 25 | 68.0% | 68.0% | 16.0% |
Cost and price | 25 | 68.0% | 68.0% | 28.0% |
Best-of recommendation | 24 | 66.7% | 66.7% | 33.3% |
Troubleshooting | 25 | 60.0% | 64.0% | 24.0% |
Trust and safety | 25 | 60.0% | 68.0% | 28.0% |
Troubleshooting is also where AI cites least. 16.0% of troubleshooting answers cite no URL at all, as do 16.0% of comparison answers. Freshness and best-of answers sit at 0.0%.
Why it matters: For new and time-sensitive questions, treat ranking as close to a requirement. For best-of, trust and troubleshooting questions, plan to earn citations beyond your own rankings too. Opttab's Prompt Volume (https://opttab.com/prompt-volume/) shows how often each topic is asked in AI search, so you can weight intents by real demand.
Finding 6: Two in five answers show a source dated this year, four in five for best-of questions
40.8% of answers show at least one cited source with a visible 2026 publication or update date (cluster 95% CI 30.6–50.8%, n=80/196).
Intent drives the share. 79.2% of best-of recommendation answers show a 2026-dated source (n=24). Cost and price answers do in 56.0%, and freshness answers in 50.0%. Comparison answers show one in 24.0%.
Answers are specific as well as dated. 88.3% name at least one product, brand or software tool (cluster 95% CI 80.1–94.9%).
AI surface | Answers | Shows a source dated 2026 | Names a product, brand or tool |
|---|---|---|---|
Bing AI Mode | 40 | 55.0% | 85.0% |
Google AI Mode | 40 | 47.5% | 87.5% |
Google AI Overviews | 37 | 40.5% | 97.3% |
Gemini | 40 | 32.5% | 85.0% |
ChatGPT | 39 | 28.2% | 87.2% |
Why it matters: Your best-of and price pages will sit next to sources that show this year's date. Put a visible, honest update date on the pages you want cited. Opttab's Content Studio (https://opttab.com/generate-content/) writes update briefs from your visibility gaps.
Also worth knowing
Most sources are American. 55.6% of the 1,204 answer-domain citation pairs point to US sites (95% CI 52.8–58.4%).
Gemini cites least. Only 28 of Gemini's 40 answers cite a source. ChatGPT cited one in 39 of 39 answers.
Google AI Overviews cites in bulk. 91.9% of Google AI Overviews list six or more URLs. Only 2.5% of Gemini answers do.
Comparisons name fewer products. 68.0% of comparison answers name a specific product, brand or tool. Trust and safety answers do in 100.0%.
Vendor content travels. After youtube.com and reddit.com, the most-cited domains are support.google.com (6.1%), Software vendor C (5.6%) and shopify.com (5.6%).
The playbook
Opttab helps 400+ brands get found and recommended by AI answer engines. Each play below comes from the findings above. To get your own baseline first, run the free AI Visibility Report (https://opttab.com/ai-visibility-report/). It needs no credit card.
1. Measure your own overlap, prompt by prompt Evidence: 68.0% of answers with sources cite a top-10 URL overall. In ChatGPT the share is 46.2%, and in Google AI Mode it is 57.9%. Do this: List the 30 to 50 questions that matter most in your category. Record Google's top 10 for each. Then check which AI surfaces cite your ranking pages and which cite someone else. In Opttab: Prompt Generator (https://opttab.com/prompt-generator/) builds the question set from your topics or your Google Search Console keywords. AI visibility tracking (https://opttab.com/ai-visibility/) then records which pages AI cites, prompt by prompt and model by model.
2. Push page-two pages onto page one first Evidence: No answer's best-cited URL ranked in positions 11 to 20 (0 of 196). 54.6% came from positions 1 to 3. Every Google AI Overview with sources cited a top-10 URL. Do this: Start with pages sitting just outside the top 10 on questions that trigger Google AI Overviews. On your most valuable questions, aim for the top three. In Opttab: GEO / AEO actions (https://opttab.com/geo-aeo/) maps each prompt to your pages and lists the content, product and on-site technical fixes. Content Studio (https://opttab.com/generate-content/) turns the prompts where you lose into briefs.
3. Work the pages that sit next to yours Evidence: Only 7.1% of answers cite top-10 pages alone. The first non-ranking source is a publisher or review site in 32.1% of answers. A brand or vendor site comes first in 24.0% (indicative). Do this: Send accurate product, pricing and spec facts to the publishers, review sites and comparison pages that cover your category. Keep your own vendor pages specific. In Opttab: Answer Engine Analytics shows the sources behind each answer, per prompt and surface. AXP bot-ready pages (https://opttab.com/axp-site/) serve AI bots a clean, structured version of your own pages and keep it in sync when they change.
4. Show up where Google AI Overviews reads the community Evidence: Google AI Overviews cites a forum or community thread in 64.9% of answers. youtube.com (18.9%) and reddit.com (14.3%) are the two most-cited domains. Do this: Answer real questions in the relevant threads. Keep product videos current, with the same facts as your site. In Opttab: Answer Engine Analytics lists the threads and videos each surface cites for your tracked prompts, so you know where to show up.
5. Date your best-of and price pages honestly Evidence: 79.2% of best-of answers show a cited source dated 2026. For cost and price answers, the share is 56.0%. Do this: Put a visible update date on buying guides and pricing pages. Change it only when the content changes. In Opttab: Content Studio (https://opttab.com/generate-content/) writes refresh briefs and updates for the best-of and pricing prompts where competitors are cited instead of you.
6. Keep two scorecards: rankings and citations Evidence: Overlap ran from 100.0% in Google AI Overviews to 46.2% in ChatGPT. Two surfaces with similar rankings can build answers from different sources. Do this: Report rankings and AI citations side by side, per surface. Re-run the same prompts monthly to see how answers move. In Opttab: AI visibility tracking (https://opttab.com/ai-visibility/) reports visibility per AI platform. Agent and website analytics (https://opttab.com/agent-analytics/) shows which AI bots visit your site, which pages they read, and the traffic and conversions AI search sends. Connect Google Search Console and Looker Studio through Integrations (https://opttab.com/integrations/). Start with the free AI Visibility Report (https://opttab.com/ai-visibility-report/), take the 7-day free trial (https://opttab.com/pricing/), or book a demo (https://opttab.com/demo-book/).
Methodology
Prompts. We wrote 40 shopper and buyer questions in English for the United States. They cover eight intents: explainer, how-to, troubleshooting, trust and safety, comparison, cost and price, best-of recommendation and freshness. Each intent has 22 to 25 answers.
Google baseline. For each question we recorded Google's US organic results for the same query on the same day. We then matched every cited source against the top 10 after URL normalization, at URL level and at domain level.
Surfaces. Every question went to five AI surfaces: ChatGPT, Gemini, Google AI Mode, Google AI Overviews and Bing AI Mode. Perplexity was planned but not collected. The cloro collection API returned MODEL_UNAVAILABLE for all 40 Perplexity requests because it could not access Perplexity in logged-out sessions. This is not a ranking of models.
Collection. We collected answers on 9 November 2026 in logged-out sessions, geo-targeted to the United States, through the cloro collection API. 200 answer slots gave 196 answers. Google AI Overviews returned no answer on 3 searches and ChatGPT on 1. 181 answers (92.3%) cited at least one source. We also classified 1,204 answer-domain citation pairs covering 824 domains by type and home market.
Coding. We coded each answer against a written ten-field codebook with model-assisted extraction. The fields cover sources, citation counts, top-10 URL and domain matches, best-cited Google position, the type of the first non-ranking source, forum citations, named products and 2026-dated sources.
Quality checks. An independent second coder re-coded 120 answers (1,200 judgements), with 82.8% overall agreement. Agreement was highest for named products (99.2%, kappa 0.961) and top-10 domain matches (95.8%, kappa 0.895). It was lowest for top-10 URL matches (71.7%, kappa 0.467) and non-ranking source type (58.3%, kappa 0.458).
Statistics. Coded headline figures use 95% Wilson intervals and a cluster bootstrap by prompt, reported as cluster 95% CI. Exact top-10 computations carry 95% intervals and lead wherever they exist. We ran no significance tests between groups. Instead, we say whether intervals overlap.
Earlier studies. Our Black Friday 2026 (31 October 2026) and Cyber Monday 2026 (4 November 2026) studies used different prompts, countries and surfaces. Any comparison with them is directional.
Limitations
One day, one run. This is a single run on 9 November 2026 with 40 questions. AI answers vary from run to run. Treat the figures as a snapshot, not a trend.
One market, one language. We asked every question in English for the United States. Other markets and languages may overlap with Google differently.
Logged-out only. Signed-in or personalised experiences may cite differently. Perplexity could not be collected at all.
Small groups. Surface groups hold 28 to 40 answers with sources, and intent groups hold 22 to 25 answers. Read differences between surfaces and intents as indicative.
Modest agreement on some fields. Top-10 URL matches (kappa 0.467) and non-ranking source type (kappa 0.458) had the weakest agreement. We lead with exact computations where they exist and call the source-type split indicative.
Observational. The study shows how often AI citations overlap with Google rankings. It does not show that ranking causes citation. A citation is also not the same as a click, traffic or a sale.
Vendor names. Competing software vendors among the cited domains are shown as Software vendor A, Software vendor B, Software vendor C; their figures are unchanged.
