Live · updated hourly

Most-cited websites by ChatGPT

The publishers and platforms ChatGPT actually reaches for when it answers, ranked by citation share, with weekly movement.

Rolling 4 weeks: 46,808 citations · 15,247 model responses · every named domain seen across ≥ 2 independent projects

Top-50 stability: 20% retained week to week: a fast-changing source list

By modelAll models Claude Perplexity Gemini Grok Google AIO Google AI Mode

This week's movers

Citation-share change, week commencing 2026-08-31 vs 2026-09-07. Minimum 25 citations in both weeks and a 0.2pp move to qualify; domains newly added to monitoring don't count as movers.

Gaining

gov.co 0.1% → 1.0% ▲ +0.9pp

gov.ae 0.2% → 0.6% ▲ +0.4pp

samsung.com 0.3% → 0.6% ▲ +0.3pp

whowhatwear.com 0.2% → 0.4% ▲ +0.2pp

Declining

moneyhelper.org.uk 2.2% → 0.7% ▼ -1.6pp

gov.uk 3.0% → 1.9% ▼ -1.1pp

which.co.uk 1.1% → 0.6% ▼ -0.5pp

europa.eu 0.6% → 0.4% ▼ -0.2pp

ChatGPT's top sources

Top 50 domains by share of all citations in the rolling window. The trend column is the weekly share trajectory.

#	Domain	Share	Citations
1	gov.uk	2.0%	914
2	moneyhelper.org.uk	1.1%	533
3	walmart.co.cr	1.1%	500
4	which.co.uk	0.8%	378
5	kia.com	0.7%	327
6	hyundai.com	0.7%	320
7	fi.cr	0.6%	279
8	techradar.com	0.6%	265
9	nhs.uk	0.5%	256
10	fca.org.uk	0.5%	243
11	apple.com	0.5%	241
12	tesla.com	0.5%	230
13	gob.pe	0.5%	224
14	baccredomatic.com	0.5%	223
15	ahorraya.app	0.5%	222
16	nih.gov	0.5%	214
17	samsung.com	0.5%	214
18	gov.ae	0.5%	211
19	eurostar.com	0.4%	202
20	bancobcr.com	0.4%	192
21	digitalcameraworld.com	0.4%	183
22	reddit.com	0.4%	177
23	europa.eu	0.4%	170
24	google.com	0.4%	166
25	rtings.com	0.3%	161
26	tomsguide.com	0.3%	157
27	citizensadvice.org.uk	0.3%	156
28	abi.org.uk	0.3%	152
29	nespresso.com	0.3%	144
30	volvocars.com	0.3%	139
31	go.cr	0.3%	136
32	gov.co	0.3%	134
33	nsinternational.com	0.3%	130
34	wikipedia.org	0.3%	130
35	dpreview.com	0.3%	130
36	financial-ombudsman.org.uk	0.3%	119
37	gob.ar	0.2%	115
38	moneysavingexpert.com	0.2%	103
39	womanandhome.com	0.2%	101
40	t3.com	0.2%	100
41	theguardian.com	0.2%	97
42	ofcom.org.uk	0.2%	97
43	whowhatwear.com	0.2%	97
44	gouv.fr	0.2%	95
45	aad.org	0.2%	95
46	whatcar.com	0.2%	93
47	gov.sa	0.2%	90
48	johnlewis.com	0.2%	90
49	ford.com.au	0.2%	85
50	runrepeat.com	0.2%	83

The race for the top

Where the leading sources sat against each other, week by week. Rank space makes every crossing an event.

Figure 1The race for the top of the index

Rank within the top 6 sources, week by week. Only the biggest mover is coloured; the greys are the rest of the field.

Data table

Series	24 Aug	31 Aug	7 Sept	14 Sept
gov.uk	3	1	1	2
moneyhelper.org.uk	5	2	3	6
walmart.co.cr	2	5	2	1
which.co.uk	4	4	4	5
kia.com	6	3	6	4
fi.cr	1	6	5	3

2 places gained: kia.com, #6 → #4 (14 Sept)

How sourced are the answers?

Citations per response over the last 12 complete weeks: whether the models are leaning on the web more or less.

Figure 2Citations per answer, week by week

Each cell is one model in one week; hotter cells mean answers leaned on more sources. Single-hue ramp; the scale runs from the lowest to the highest cell shown.

Data table

Series	22 Jun	29 Jun	6 Jul	13 Jul	20 Jul	27 Jul	3 Aug	10 Aug	17 Aug	24 Aug	31 Aug	7 Sept
ChatGPT	3	2.7	3	3.1	2.6	2.8	5	3.7	2.7	3.3	3.1	2.7

Methodology

What's counted: every citation/source URL ChatGPT returned across the rolling 4 ISO weeks, aggregated from every analysis run on BrandInsightAI. Citations deduplicate to one registrable domain per model response.
Quality gate: a domain must appear across ≥ 2 distinct projects to be named; shares are computed against ALL citations in scope, so the gate hides names, never distorts percentages.
Movers: week-over-week comparisons use the two most recent COMPLETE ISO weeks (2026-08-31 → 2026-09-07); the in-progress week is excluded because it refreshes hourly, and a domain must clear the citation floor in BOTH weeks, so a newly monitored category entering the dataset never shows up as a gainer.
Freshness: the snapshot behind this page recomputes every hour. All figures are measured, never estimated.

Where next

This board answers "which websites does AI cite?". These answer the questions that come next — which brands it recommends, which sources each model prefers, and how any of it is counted.

By modelAll models Claude Perplexity Gemini Grok Google AIO Google AI Mode

Explorer The AI Source Index Dig into which sources each model trusts, filterable by market — compiled into a quarterly report you can download as PowerPoint or PDF. Updated hourly Live leaderboards The Category Index The board behind the story — every category ranked: brand visibility, citation share and how often a read page actually gets cited. Refreshed weekly Live leaderboard The AI Desire Index What AI assistants want from a brand: attraction, affinity and attachment for every brand in every category, indexed and refreshed weekly. Refreshed weekly Journal Analysis & articles The data journalism series — weekly citation movers, model vs model, what gets cited and why. Every edition reviewed before it publishes. New articles weekly Data story The Category Report How AI shops for Britain: the brands ChatGPT, Claude and Gemini actually recommend across 20 consumer categories, told as an animated story. Refreshed weekly Reference Methodology How every number is measured: real AI answers, counted not estimated, with sample sizes and limitations stated on every page. The rules behind it all Start here Insights overview The whole picture in one page: every live board, every data story and every published analysis, with what each one answers. The map of everything