Does AI cite sources when it criticizes a brand?
Usually not. Of 64,899 times AI criticized a ranked brand, 42,748, or 65.87%, had no cited page attached to that brand in the same answer.
By Dimitry Apollonsky · August 13, 2026 · 12 min read
Contents
- Nearly two in three brand criticisms had no source attached
- Criticism was less likely to carry a citation than other brand language
- The engine split was only 2.09 percentage points
- First-place criticism was cited more often than lower-ranked criticism
- Changing the language definition did not remove the gap
- Only 3.67% of criticisms had four or more attached sources
- Industry citation rates differed by 15.89 percentage points
- Atlassian and Datadog had the most uncited criticisms
- The 34.13% citation rate held under stricter matching
- What marketers should do
- Get the data
- Sources
- Related research
In one observed cut of the Parse mirror, we analyzed 940,245 reviewed AI statements across 900,065 ranked brand appearances, 74,659 brands, 224,593 answers, and 16,161 organic prompts on ChatGPT Search and Google AI Mode from May 24 through July 16, 2026.
Nearly two in three brand criticisms had no source attached
AI made a specific negative claim in 64,899 ranked brand appearances. In 42,748 of them, or 65.8685%, no cited page was attached to that brand in the same answer. The remaining 22,151 had at least one attached cited page.
A criticism audit should start with the uncited cases. An attached page still requires review because attachment does not prove that the page supports the negative claim.
Takeaway
Criticism was less likely to carry a citation than other brand language
A cited page was attached to 22,151 of 64,899 criticized brand appearances, or 34.1315%. Other ranked brand appearances had an attached cited page in 366,725 of 835,166 cases, or 43.9104%. Criticism was cited 9.7789 percentage points less often.
Overall citation coverage is not a sufficient evidence check. Separate negative claims from other brand language when deciding which answers need source review.
The engine split was only 2.09 percentage points
ChatGPT Search attached a cited page to 18,261 of 54,056 criticized brand appearances, or 33.7816%.
Google AI Mode did so for 3,890 of 10,843, or 35.8757%.
Google AI Mode was higher by 2.0941 percentage points.
The uncited-criticism problem appeared on both engines. Audit both, but do not treat the small aggregate difference as evidence that one engine is more accurate.
First-place criticism was cited more often than lower-ranked criticism
AI attached a cited page to 3,307 of 6,610 first-place criticisms, or 50.0303%. The rate was 36.0515% at second or third and 30.2821% at fourth or lower. First place exceeded fourth or lower by 19.7482 percentage points.
Recommendation rank changes the citation baseline. Compare criticism at similar ranks before deciding that one brand or prompt has an unusual evidence gap.
Takeaway
Changing the language definition did not remove the gap
Citation attachment was 34.1315% for specific criticism, 26.8122% for negative overall tone, 35.3872% for reluctant recommendations, and 30.0071% for explicit rejection. Each definition left most affected brand appearances without an attached cited page.
A general sentiment score cannot replace claim-level review. Track the exact type of negative language because criticism, reluctance, and rejection describe different decisions.
| Reluctant recommendation | 46,401 | 16,420 | 35.3872% |
| Specific criticism | 64,899 | 22,151 | 34.1315% |
| Explicit rejection | 4,239 | 1,272 | 30.0071% |
| Negative overall tone | 9,574 | 2,567 | 26.8122% |
Only 3.67% of criticisms had four or more attached sources
Among 64,899 criticized brand appearances, 42,748 had no attached cited page, 12,431 had one, 7,341 had two or three, and 2,379 had four or more. Four or more sources accounted for 3.6657% of the total.
Source count helps prioritize review, but it is not a support score. One relevant page can be stronger evidence than several pages that do not contain the claim.
Industry citation rates differed by 15.89 percentage points
Among displayed industries with at least 500 criticized brand appearances, Sales and Marketing attached a cited page in 398 of 1,033 cases, or 38.5286%. Blockchain and Cryptocurrency did so in 134 of 592, or 22.6351%. The difference was 15.8935 percentage points.
Use an industry baseline before calling a brand's citation rate unusual. The comparison identifies where review may be useful and does not show that industry caused the difference.
| Sales and Marketing | 1,033 | 398 | 38.5286% |
| Education | 887 | 329 | 37.0913% |
| Artificial Intelligence | 2,388 | 801 | 33.5427% |
| Financial Services | 4,387 | 1,422 | 32.4140% |
| Software | 4,906 | 1,325 | 27.0077% |
| Hardware | 1,097 | 296 | 26.9827% |
| Blockchain and Cryptocurrency | 592 | 134 | 22.6351% |
Atlassian and Datadog had the most uncited criticisms
Atlassian had 635 uncited criticisms among 1,000 criticized brand appearances. Datadog had 548 among 915. Alphabet,
Notion,
ClickUp, Grafana, Elastic,
Microsoft,
Asana, and Cisco completed the top ten by uncited-criticism count.
This is a review-volume leaderboard, not a brand-quality ranking. Compare each brand's prompt mix, recommendation rank, and cited pages before drawing a reputation conclusion.
Takeaway
The 34.13% citation rate held under stricter matching
The main citation rate was 22,151 of 64,899, or 34.1315%. A June 1 start returned 22,149 of 64,896, or 34.1300%. Exact brand identities returned 22,074 of 64,899, or 34.0129%. Requiring higher-confidence source attachment returned the same 34.1315% main rate.
The study counted one brand family once per answer and excluded language that was not validated, brands without a ranked position, and 123 brand-source links without an identifiable cited page. It measures citation attachment, not whether a source mentions the brand, supports the criticism, or makes the criticism accurate.
What marketers should do
No cited page was attached in 42,748 of 64,899 observed brand criticisms. Citation attachment varied more by recommendation rank and industry than by engine, and only 3.6657% of criticisms had four or more attached sources.
Track specific criticism separately from general sentiment. Audit uncited criticism first. For cited criticism, open the attached page and verify whether it contains and supports the claim. Compare brands at similar ranks within the same industry. Repeat the fixed-window method next quarter before calling a difference movement.
Get the data
Sources
- Semrush: Why 62% of AI citations don't lead to brand mentions · accessed 2026-08-13
- Seer: LLM ghost citations · accessed 2026-08-13
- BrightEdge: When AI goes negative · accessed 2026-08-13
- Evaluating verifiability in generative search engines · accessed 2026-08-13