Airlines 🇬🇧
British Airways led with 44% recommendation rate (24/55 answers).
55 eligible answers (72 captured, excluding 17 No clear answer) · 17 brands
Seenford research
Seenford AI Recommendation Index
Edition 1 · published 2 November 2026 · 1,223 answers · 8,835 sources · 6 engines · 5 categories
Fewer than 1 in 10 sources AI cites are brands' own websites.
What six AI engines say about brands across five categories: who they recommend, what they cite and where brands' own websites stand.
About the Index
The Seenford AI Recommendation Index shows what six AI engines (ChatGPT, Google Gemini, Anthropic Claude, Perplexity, Google AI Overviews and Microsoft Copilot) say when people ask them about brands: which brands they mention, which they recommend, and which sources they rely on. A new edition is published every quarter, using the same method each time, so changes can be compared from one edition to the next. We publish it because more people now ask AI engines for advice before they buy, yet most brands cannot see how those answers describe them. Tracking this is often called AI visibility, or answer engine optimisation (AEO). The Index gives marketing teams an independent, evidence-based view of how their category is being retold by AI.
Published every quarter by Seenford · seenford.com
Key findings
- 01Share
8%
Brands' own websites made up 8% of the sources AI engines cited.
- 02Share
9%
Forums and social platforms accounted for 9% of cited sources.
- 03Share
7.2
AI engines cited an average of 7.2 sources per answer.
- 04Share
57%
Recommended made up 57% of brand mentions across the five categories.
- 05Share
10%
On brand-name questions, cautions made up 10% of mentions.
Figures describe the sources AI engines cited, not website traffic. Answers from 2026-09-23 to 2026-10-23, counted once per question, engine and week.
The headline data
What AI reads
Blog and guide
22%
1,946 sources
News and media
19%
1,635 sources
Other / unknown
14%
1,251 sources
Review and consumer
12%
1,086 sources
Retailer and marketplace
9%
799 sources
Forum and social
9%
750 sources
Brand's own site
8%
698 sources
Comparison and booking
7%
580 sources
Government and regulator
1%
90 sources
Share of cited sources by type · Sources v1 · 2026-09-23 to 2026-10-23. Shares add up to 100%. Shares describe cited sources, not traffic.
Engines compared
ChatGPT
201 answers
Brand-owned share
22%
Forum and social share
1%
Sources per answer
5.3
Claude
203 answers
Brand-owned share
2%
Forum and social share
3%
Sources per answer
7.4
Gemini
205 answers
Brand-owned share
7%
Forum and social share
14%
Sources per answer
7.0
Perplexity
205 answers
Brand-owned share
7%
Forum and social share
5%
Sources per answer
10.9
Copilot
205 answers
Brand-owned share
5%
Forum and social share
0.4%
Sources per answer
6.6
Google AI Overviews
204 answers
Brand-owned share
9%
Forum and social share
29%
Sources per answer
6.1
Brand-owned share and forum and social share of cited sources, and sources cited per answer, for each of the 6 engines. Sources-per-answer bars are scaled to the highest engine.
Mentioned compared with recommended
Airlines (UK)
67%
72 answers across all Seenford sources
Credit cards (UK)
41%
54 answers across all Seenford sources
Supermarkets (UK)
58%
82 answers across all Seenford sources
Fast food (US)
51%
54 answers across all Seenford sources
Car insurance (US)
65%
54 answers across all Seenford sources
Recommended as a share of mentioned brand readings, by category. Recommended = Top pick + Recommended.
Caution rate by question type
Best overall
2%
358 answers
Brand check
49%
309 answers
Comparison
2%
59 answers
Service and reliability
0%
60 answers
Specific need or audience
5%
359 answers
Value and price
5%
42 answers
Cautioned readings as a share of mentions, by question type.
Explore the categories
Recommendation rate = answers where the brand was recommended (Top pick or Recommended) ÷ eligible answers in the category, counted once per question, engine and week. Eligible answers leave out failed answers and No clear answer results. Category reports may show the same run on a mentions basis.
Credit cards 🇬🇧
Lloyds led with 37% recommendation rate (19/52 answers).
52 eligible answers (54 captured, excluding 2 No clear answer) · 15 brands
Supermarkets 🇬🇧
Tesco led with 42% recommendation rate (29/69 answers).
69 eligible answers (72 captured, excluding 3 No clear answer) · 13 brands
Fast food 🇺🇸
Chick-fil-A led with 57% recommendation rate (29/51 answers).
51 eligible answers (54 captured, excluding 3 No clear answer) · 13 brands
Car insurance 🇺🇸
USAA led with 74% recommendation rate (39/53 answers).
53 eligible answers (54 captured, excluding 1 No clear answer) · 13 brands
Key actionable advice
What brands should do
- Ask the engines the questions your customers ask, with your brand name and without it.
- Look at what the engines cite, not just what they say.
- Check whether the pages AI reads are yours or somebody else’s.
- Fix what the engines read first, then measure again.
- Track the same questions weekly, so movement is real and not luck.
- Treat one-off answers as evidence, not as a trend.
See what AI says about your brand
Get a free one-off reading across the same engines, or sign up and we will email you when the next edition is published.
Method and limitations
Six engines: ChatGPT, Google Gemini, Anthropic Claude, Perplexity, Google AI Overviews and Microsoft Copilot. Answers were captured from 2026-09-23 to 2026-10-23 in the UK and US, in English.
Each question counts once per engine and week: the latest captured answer in that week.
Top-level figures combine all answers Seenford collected (research runs, Quick Reports and tracking), anonymised, with no customer brands named. Category pages use dedicated research runs only, where brands are named on purpose.
Shares describe cited sources, not traffic or influence.
Verdicts: Top pick (first recommendation in an unbranded answer), Recommended, Mentioned in passing, Named with caution, Not mentioned. Failed answers and No clear answer results are excluded from rates.
Each category page is one research run, so it is a snapshot, not a trend.
The full methodology explains how Seenford takes each sample.