How often AI recommends a business: 2026 data
When a buyer asks AI who to buy from, the business they are asking about is named in 35.2% of answers, recommended in 20.6%, and its own website is cited as a source in 0.6%. From 7,552 answers by five engines, measured by Citeon AI.
- Named in 35.2% of answers, recommended in 20.6%: being known is not being chosen.
- The business's own site is a cited source in 0.6% of answers. The engines read reddit.com, en.wikipedia.org and youtube.com instead.
- Asked the same question on the same day, two engines name the same business first 26 to 35.7% of the time.
- Of 1,351 questions put to four or more engines, 53.3% got no recommendation from any engine and 8% from four or more.
The headline numbers
Each answer is one engine's reply to one buyer question about one business on one day. A question is something a customer would type, like "best accountant in Rotterdam" or "which tool tracks what ChatGPT says about my brand". The three rates are scored on every answer: did the answer name the business, did it present the business as the pick, and did it list the business's own website among its sources.
By engine
| Engine | Answers | Business named | Recommended | Own site cited |
|---|---|---|---|---|
| Perplexity | 1,777 | 32.2% | 20.1% | 0.6% |
| ChatGPT | 1,776 | 31.8% | 21.6% | 0.3% |
| Claude | 1,687 | 33.8% | 18.8% | 0.7% |
| Gemini | 1,543 | 40.7% | 26.1% | 0.5% |
| Google AIO | 769 | 42.4% | 12.7% | 1.2% |
Every answer measured, 24 August 2026 to 7 October 2026. Google AI Overviews is asked on the Optimize plan only, so its sample is smaller.
Google AI Overviews names a business most often and recommends one least often: it summarises a list, it rarely picks. Gemini both names and recommends most among the chat engines. ChatGPT, Claude and Perplexity sit within a few points of each other on naming and spread wider on recommending.
Where a named business sits
An answer is short. With about 7 brands per answer and no page two, a business is either in the first few names or absent. When it is named, it is first in 45.1% of those answers, which says the engines do have a favourite more often than not.
Where the engines get their answers
The engines list the pages they used. Counting each domain once per answer, and never counting a business's own site as a source for that business, these are the platforms cited across the most answers. A platform is listed only if answers about five or more different businesses cite it, so no single business's competitor appears here.
| Platform | Answers citing it | Share of answers with sources | Businesses whose answers cite it |
|---|---|---|---|
| reddit.com | 1,321 | 19.1% | 32 |
| en.wikipedia.org | 767 | 11.1% | 23 |
| youtube.com | 436 | 6.3% | 15 |
| linkedin.com | 293 | 4.2% | 20 |
| github.com | 270 | 3.9% | 7 |
| finance.yahoo.com | 257 | 3.7% | 7 |
| zapier.com | 253 | 3.7% | 9 |
| medium.com | 218 | 3.2% | 13 |
| apps.apple.com | 188 | 2.7% | 6 |
| facebook.com | 185 | 2.7% | 14 |
| forbes.com | 165 | 2.4% | 13 |
| amazon.com | 144 | 2.1% | 8 |
| dev.to | 139 | 2% | 7 |
| techradar.com | 125 | 1.8% | 7 |
| instagram.com | 115 | 1.7% | 8 |
6,916 answers listed at least one source.
The top sources, per engine
| Engine | Most cited platforms, in order |
|---|---|
| ChatGPT | reddit.com, en.wikipedia.org, github.com, checkatrade.com |
| Perplexity | reddit.com, linkedin.com, finance.yahoo.com, en.wikipedia.org, zapier.com |
| Claude | en.wikipedia.org, github.com, finance.yahoo.com, medium.com, dev.to |
| Gemini | reddit.com, en.wikipedia.org, youtube.com, facebook.com, medium.com |
| Google AIO | youtube.com, reddit.com, facebook.com, instagram.com, en.wikipedia.org |
The engines do not read the same web. ChatGPT and Gemini lean on Reddit; Claude on Wikipedia and developer sites; Perplexity on Reddit, LinkedIn and finance pages; Google AI Overviews on YouTube, Reddit, Facebook and Instagram, the only engine that puts social platforms first. The 2026 citation studies find the same split from much larger samples: about 11% of domains are cited by both ChatGPT and Perplexity.
How much the engines agree
We put the same question about the same business to several engines on the same day and compared the brands they named.
| Engine pair | Same brand first | Any of the top three shared | Questions compared |
|---|---|---|---|
| Claude and Gemini | 35.7% | 63.5% | 241 |
| Claude and Perplexity | 32.4% | 66.8% | 253 |
| ChatGPT and Claude | 32.3% | 60.8% | 260 |
| ChatGPT and Perplexity | 32.1% | 56.6% | 274 |
| Gemini and Perplexity | 32% | 59.5% | 247 |
| ChatGPT and Gemini | 26% | 55% | 258 |
Pairs where both engines named at least one brand for the same question on the same day.
| Engine pair | Cited domains shared |
|---|---|
| ChatGPT and Claude | 9.1% |
| ChatGPT and Gemini | 10.1% |
| ChatGPT and Google AIO | 9.3% |
| ChatGPT and Perplexity | 9.5% |
| Claude and Gemini | 13.9% |
| Claude and Google AIO | 12.7% |
| Claude and Perplexity | 17.2% |
| Gemini and Google AIO | 16.5% |
| Gemini and Perplexity | 13.1% |
| Google AIO and Perplexity | 9.3% |
Share of all domains either engine cites, over the whole period, that both engines cite.
Of the 1,351 questions asked to four or more engines on the same day, 53.3% ended with no engine recommending the business, 19.6% with one engine, 19.1% with two or three, and 8% with four or more. A business that checks one engine sees one fifth of its picture.
By country
| Country of the question | Answers | Business named | Recommended |
|---|---|---|---|
| Not recorded (older runs) | 5,597 | 35.9% | 23.8% |
| US | 1,153 | 45.4% | 16.8% |
| NL | 742 | 17.5% | 4% |
| GB | 60 | 1.7% | 1.7% |
The country is the market the question was asked for. Runs before the country field existed are shown as not recorded, not folded into a country.
The gap between the US and the Netherlands is a sample gap as much as a market gap: the Dutch answers are mostly local-service questions, where the engines name many businesses and pick few. Read the per-engine table for the shape that holds across countries.
How this is measured
- Each business gets a set of buyer questions, written from its site and market and organised by intent (discovery, comparison, recommendation, local).
- Every day the questions go to the engines the business tracks: ChatGPT, Gemini, Claude, Perplexity and, on the Optimize plan, Google AI Overviews. Questions rotate in halves, so each is asked every two days.
- Each answer is stored in full and scored: named, position among the brands, recommended, own site cited, and the list of source domains.
- This page reads every scored answer with one script and prints the counts. The script runs monthly; the dates above are the dates of the data.
- The data is published under CC BY 4.0: reuse the numbers and the tables with a link to this page.
What this is not: a survey, an estimate or a crawl of homepages. It is the engines' own answers, read one by one. Sample limits are stated next to every table.
Guides built on this data
- GEO vs SEO vs AEO: the differences
- What is generative engine optimization (GEO)?
- What is answer engine optimization (AEO)?
- Google AI Overviews: what they are and how to appear in them
- ChatGPT SEO: how ChatGPT chooses its sources
- What is AI search optimization?
- What is AI visibility?
- What is llms.txt?
The measures in the product: AI visibility tool, share of voice, AI Overviews tracker.
Questions people ask
What counts as recommended?
An answer counts as naming a business when the business appears in the answer text (judged by a model and checked against the brand's aliases). It counts as recommending it when the answer presents it as the pick, the first choice or the one to go with, not merely as one option in a list. Both are scored per answer, per engine, per day.
Why is the own-site citation rate so low?
Because the engines build answers from third-party pages and then name the businesses those pages talk about. In this sample a business's own website is cited as a source in 0.6% of answers, while the business is named in 35.2%. The most cited sources are platforms: reddit.com, en.wikipedia.org, youtube.com, linkedin.com.
Is this sample representative of every market?
No. It is every answer Citeon AI measured for 35 businesses between 24 August 2026 and 7 October 2026, weighted toward hospitality and local services in the Netherlands and software in the United States. The country split is on this page. Use it as a measured baseline, not as a market average.
How often is this page updated?
Monthly, by a script that reads every measured answer and rewrites the figures. The dates and counts on the page are the dates and counts of the run. Nothing is edited by hand.
Can I use these numbers?
Yes, with a link to this page and the date. If you need a cut by engine, country or intent that is not here, write to hello@citeon.ai.
Sources
- Citeon AI, prompt_results. every answer collected 24 August 2026 to 7 October 2026; script scripts/data/ai-recommendations.py in the Citeon repository
- Google Search Central, AI features and your website. how AI Overviews and AI Mode choose and show links
- 5WPR, The state of AI citations 2026. the external studies this page is compared with