Where AI Citations Actually Come From: Our Monitoring Data

The finding

Across the AI answers we monitored, the sources cited most often are platforms and publishers, not brands' own websites.

Data re-extracted and frozen 16 October 2026.

Top 12 source domains of 6,597 total citations (frozen 16 Oct 2026).

Source domain	Citations	Engines
youtube.com	500	4
reddit.com	377	3
forbes.com	117	3
techradar.com	99	3
linkedin.com	55	3
whathifi.com	48	3
finance.yahoo.com	44	1
bankrate.com	44	3
zapier.com	41	5
pcmag.com	39	2
nerdwallet.com	39	2
kiplinger.com	38	2

Method

• Universe: the brands monitored in this platform - 10 brands, 71 monitored queries.
• Engines: ChatGPT, Gemini, Google AI Overviews, Perplexity, Microsoft Copilot (the five that returned data).
• Window: 13 Aug 2026 to 30 Sep 2026.
• Measure: citations the engines made to a publication (classified as info_site), grouped by registered domain.
• Sample: 6,597 cited sources in total (frozen 16 Oct 2026).
• Excluded: our own domains and test domains.

Platform-count note: this run covers the five engines that produced data. We do not run every engine on every query, and we do not claim eight.

Limitations

• This is our measurement, not a survey of the AI industry.
• The monitored brands are a mix of internal, pilot and demo accounts. The AI answers are real; the brand set is not a representative sample of all categories.
• A citation is not an endorsement, and a citation is not a recommendation.
• Counts are citations to a domain, not unique pages.
• The window is roughly seven weeks, not a year.
• The published total is a snapshot taken 30 Sep 2026. Citation rows are periodically reprocessed, so a live re-run returns a comparable but not identical figure (about 6,600 today).

How we computed it

This is our internal query, published so the method can be audited rather than asserted. It reads our own measurement database, which readers do not have access to, so treat it as documentation rather than something you can run yourself. The figures live in the machine-readable file linked above.

SELECT LOWER(TRIM(LEADING 'www.' FROM SUBSTRING_INDEX(SUBSTRING_INDEX(Url,'://',-1),'/',1))) AS domain,
                       COUNT(*) AS cites,
                       COUNT(DISTINCT Platform) AS engines
                FROM tracker_citations
                WHERE CitationSourceType = 'info_site'
                  AND Url LIKE 'http%'
                  AND Date >= '2026-08-13' AND Date < '2026-10-01'
                  AND TenantId <> 9001
                  AND Url NOT LIKE '%aeospotlight.com%'
                  AND Url NOT LIKE '%testinc.com%'
                  AND Url NOT LIKE '%farkasfloors.com%'
                GROUP BY domain
                ORDER BY cites DESC;
                

The published table is the snapshot taken 30 Sep 2026; re-running today returns a comparable figure, not an identical one.

The machine-readable version of this dataset is at /data/ai-citation-source-domains-2026-09/dataset.json.

What it suggests

If the sources AI cites are mostly third-party platforms rather than brand websites, then being cited depends substantially on whether third parties cover you, not only on what your own site says.

Related

• Where AI citations come from: the research
• Does schema markup get you cited?

## Read the analysis The numbers, the why and what to do about them are broken down in plain language here: **/data/what-gets-cited-in-ai-2026/**. The two pages are the measurement and its interpretation — use whichever fits.