Does AI praise and criticize the same brand?
Sometimes. Of 1,564,880 times a brand appeared in an AI recommendation list, 112,376 included both praise and criticism about that brand.
By Dimitry Apollonsky · August 22, 2026 · 11 min read
Contents
- Praise and criticism appeared together in 7.2% of ranked brand appearances
- Most criticism appeared beside praise
- ChatGPT Search mixed praise and criticism five times as often as Google AI Mode
- Mixed language was 3.2 points more common below first place
- The praise and criticism usually occurred in one statement
- Alphabet, Atlassian, and Microsoft had the most mixed appearances
- Industry rates differed by almost six percentage points
- The result held under narrower counting rules
- What marketers should do
- Get the data
- Sources
- Related research
In the Parse index, we analyzed 1,650,466 reviewed brand statements across 1,564,880 ranked brand appearances, 390,170 AI answers, 130,732 brands, and 17,320 organic prompts on ChatGPT Search and Google AI Mode from May 24 through August 19, 2026.
Praise and criticism appeared together in 7.2% of ranked brand appearances
AI used both praise and criticism for the same brand in 112,376 of 1,564,880 ranked brand appearances, or 7.2%. A ranked brand appearance is one consolidated brand with a direct recommendation rank and reviewed language in one observed AI answer.
BrightEdge's closest study counts prompts where an answer praised some brands while criticizing others. This study counts one ranked brand receiving both, so the measurements are different. A favorable or negative summary can hide a qualified recommendation. Review the claims attached to the brand before treating one sentiment label as the answer's complete judgment.
Takeaway
Most criticism appeared beside praise
Among 132,735 ranked brand appearances with criticism, 112,376 also contained praise, or 84.7%. The remaining 20,359, or 15.3%, contained criticism without praise.
A criticism count is not a count of brands that AI recommended against. Separate criticism from criticism that appears inside an otherwise positive assessment.
ChatGPT Search mixed praise and criticism five times as often as Google AI Mode
ChatGPT Search mixed the two in 93,557 of 776,803 ranked brand appearances, or 12.0%. Google AI Mode did so in 18,819 of 788,077, or 2.4%. The ChatGPT Search rate was 5.0 times as high, a 9.7-point difference.
Keep separate engine baselines. A blended sentiment rate hides the largest split in this study and does not show that either engine is more accurate.
Takeaway
Mixed language was 3.2 points more common below first place
Mixed language appeared in 13,037 of 277,988 first-place appearances, or 4.7%; 36,091 of 480,522 second- or third-place appearances, or 7.5%; and 63,248 of 806,370 appearances at fourth or lower, or 7.8%. Fourth or lower exceeded first place by 3.2 points.
Compare brands at similar recommendation ranks. The pattern does not show that rank caused the language, but it makes rank necessary context for a sentiment audit.
The praise and criticism usually occurred in one statement
At least one reviewed statement itself contained both a positive and negative claim in 111,780 of 112,376 mixed appearances, or 99.5%. Separate statements supplied the two sides in 596 appearances, or 0.53%.
Answer-level sentiment alone is too coarse for this pattern. Keep each positive and negative claim tied to the sentence and brand it describes.
Alphabet, Atlassian, and Microsoft had the most mixed appearances
Alphabet had 1,363 mixed appearances among 19,254 analyzed, or 7.1%. Atlassian had 1,218 of 6,742, or 18.1%. Microsoft had 1,168 of 16,919, or 6.9%. Seven more cleaned brand names complete the leaderboard.
This is a review-volume list, not a brand-quality ranking. Use the count to prioritize answers, then compare rates, prompts, recommendation rank, and exact claims.
Takeaway
Industry rates differed by almost six percentage points
Among industries with at least 20,000 analyzed appearances, Software recorded 8,690 mixed appearances among 78,780, or 11.0%. Consumer Goods recorded 1,056 among 20,803, or 5.1%. The difference was 6.0 percentage points.
Use an industry baseline before calling a brand's mixed-language rate unusual. The result describes the observed prompt corpus and does not show that industry caused the difference.
The result held under narrower counting rules
The main rate was 112,376 of 1,564,880, or 7.2%. A June 1 start returned 109,945 of 1,546,306, or 7.1%. Complete weeks through August 16 returned 112,375 of 1,564,826, or 7.2%. Exact brand identities returned 112,404 of 1,588,374, or 7.1%.
The result was not determined by the late-May start, trailing partial dates, or brand-family consolidation. The study excludes language that could not be tied to a stable brand identity and direct recommendation rank. It does not measure factual accuracy, source support, causal drivers, buyer opinion, or user behavior.
What marketers should do
Mixed language appeared in 112,376 ranked brand appearances. ChatGPT Search's rate was 5.0 times Google's, and fourth-or-lower appearances exceeded first place by 3.2 points.
Track positive and negative claims separately. Preserve the engine, prompt, brand, and recommendation rank with each claim. Review high-volume mixed cases first. Check source support and factual accuracy before changing messaging. Repeat the fixed-window method next quarter before calling a difference movement.
Get the data
Sources
- BrightEdge: When AI goes negative · accessed 2026-08-22
- Semrush: AI Visibility Brand Performance Reports · accessed 2026-08-22
- G2: The Answer Economy, 2026 AI Search Insight Report · accessed 2026-08-22
- ACL: Recognizing conflict opinions in sentiment classification · accessed 2026-08-22