The AI engine report card: ChatGPT Search vs Google AI Mode

The report card is ten metrics computed per engine on one matched-prompt panel: the same 17,083 organic prompts, answered by both ChatGPT Search and Google AI Mode between May 24 and July 19, 2026. Because both engines answer the same questions, every gap in the scorecard reflects engine behavior, not question mix.

The three metrics about which brands to name are close: brand naming rate differs by 1.9%, clear-#1 rate by 3.1%, and brands per answer by 15%. The seven metrics about how the engine behaves — citations per answer, citation-free answers, source diversity, owned-domain citations, hedging, sentiment, and explicit negatives — split by factors of 2.7 to 36. The engines roughly agree on the shortlist and disagree on almost everything around it.

Every metric below is computed on the matched-prompt panel in the single window. The gap column divides the larger value by the smaller, so 1.0 means a tie. The rest of this article walks through the rows one at a time.

ChatGPT Search returned an answer with no citations at all in 40,224 of 224,898 answers, or 17.9%. Google AI Mode did so in 1,086 of 218,356 answers, or 0.50%. That is a 36-times gap — the largest on the report card.

Roughly one in six ChatGPT Search answers names and ranks brands with no cited source to influence. Google AI Mode almost always attaches sources. This is an observed difference in what each engine exposes, not a claim about what it retrieved internally.

Google AI Mode attached 3,048,955 citations across its 218,356 panel answers, an average of 14.0 per answer (median 12). ChatGPT Search attached 943,285 across 224,898 answers, an average of 4.2 (median 3).

Semrush's 2026 index of 126 million AI search prompts reports the opposite ordering — ChatGPT at 15.4 sources per response and Google AI Mode at 11.4 — on a different prompt set, window, and counting method. Superlines' 2026 platform comparison matches our direction, describing Google AI Mode at 5-10+ sources against 3-8 for ChatGPT. Our figures are what one matched-prompt panel exposed in one window; the disagreement across studies is itself evidence that citation volume is unstable across prompt mixes and engine builds, so measure it on your own prompts.

ChatGPT Search named at least one brand directly in 207,306 of 224,898 answers, or 92.2%. Google AI Mode did in 205,131 of 218,356, or 93.9%. On the same questions, the two engines are nearly identical on whether to bring brands into the answer at all.

Among brand-naming answers, ChatGPT Search named an average of 6.0 distinct brands (1,240,279 brand appearances across 207,306 answers) against 5.2 on Google AI Mode (1,065,265 across 205,131). The median is 5 on both; the 90th percentile is 10 brands on ChatGPT Search and 9 on Google AI Mode.

A 15% gap in shortlist size is the largest of the three brand-choice metrics, and still small next to the behavior rows. The shortlist a buyer sees is about the same length on either engine.

A clear #1 is an answer where the engine put some brand in its first recommendation slot. ChatGPT Search produced one in 165,863 of 207,306 brand-naming answers, or 80.0%; Google AI Mode in 169,232 of 205,131, or 82.5%. The share of answers with any ranked recommendation is 91.4% and 91.7%.

Neither engine is the hedging engine at the level of answer structure. Both commit to an ordered pick in most answers; the differences appear in the language wrapped around the pick, covered below.

A citing answer on Google AI Mode drew on 11.5 distinct root websites on average; on ChatGPT Search, 4.31. Within a single answer, ChatGPT Search repeats a website slightly less relative to its volume: 0.89 distinct websites per citation against 0.83.

Engine-wide, the picture inverts. Google AI Mode routed 19.9% of all its citations (605,854 of 3,044,065) to its 10 most-cited websites; ChatGPT Search routed 12.3% (116,407 of 943,285). Each Google AI Mode answer is broad, but the breadth keeps landing on the same large platforms.

Only Reddit appears in both engines' top five. ChatGPT Search's most-cited websites are Reddit, TechRadar, arXiv, Forbes, and Microsoft. Google AI Mode's are YouTube, Google's own properties, Reddit, Medium, and LinkedIn — three of its top five are social or video platforms, and its single largest destination is YouTube.

A fuller view of this comparison, at the individual-source grain, is in our study of ChatGPT Search vs Google AI Mode sources, where the two engines shared only 6.2% of cited sources per prompt.

An owned citation is a citation whose website belongs to a brand named directly in the same answer, matched on registrable root domains. On ChatGPT Search, 452,160 of 943,270 domain-resolved citations were owned, or 47.9%. On Google AI Mode, 467,473 of 3,044,065, or 15.4% — a 3.1-times gap.

Per answer, the order flips: 72.3% of Google AI Mode's brand-naming citing answers reached at least one owned website (147,654 of 204,138), against 61.5% on ChatGPT Search (104,750 of 170,233). Google AI Mode cites so many sources that a brand's own site usually appears somewhere, but as a small share; ChatGPT Search's shorter citation lists are dominated by the brands themselves.

A reluctant recommendation is a validated recommendation observation where the engine attached a qualifier — a condition, a fallback framing, an unfavorable comparison, a budget-only framing, or a risk warning. ChatGPT Search attached one to 84,778 of 830,985 recommendation observations, or 10.2%. Google AI Mode attached one to 28,459 of 879,359, or 3.2%.

Conditions lead on both engines: 61.5% of ChatGPT Search's reluctant recommendations are conditional (52,157 of 84,778) against 42.9% on Google AI Mode (12,211 of 28,459). Our corpus-wide study of recommendations with reservations found about one in ten overall; the report card shows that average is really a high-hedging engine and a low-hedging engine blended together.

Across sentiment-labeled brand-language observations on the panel, Google AI Mode was positive 77.9% of the time (734,686 of 943,022) against 59.8% on ChatGPT Search (558,353 of 933,832). ChatGPT Search shifts the difference mostly into neutral description (32.8% vs 20.1%) and into cautionary language — hedged, mixed, or negative — at 7.4% vs 1.9%, a 3.8-times gap.

A per-answer check gives the same picture: averaging positive share within each answer first yields 61.5% for ChatGPT Search and 77.9% for Google AI Mode. The gap is not driven by a few dense answers.

ChatGPT Search explicitly advised against a brand in 6,826 of 933,832 brand-language observations, or 0.73%. Google AI Mode did in 2,497 of 943,022, or 0.26%. Advice against any brand is rare on both engines, but ChatGPT Search is measurably more willing to say it.

Combined with the hedging and sentiment rows, a consistent profile emerges in this window: ChatGPT Search qualifies, neutralizes, and occasionally rejects; Google AI Mode endorses.

All metrics come from one matched-prompt panel — 17,083 organic prompts answered by both engines — over one window, May 24 through July 19, 2026, chosen because the brand-language layer is complete only through July 19. Brand metrics count direct brand mentions consolidated to canonical brands; the clear-#1 metric uses the engine's explicit recommendation ranking, which these two engines expose in this window; citation metrics count the cited-source lists attached to each answer, with websites consolidated to registrable root domains; hedging and sentiment come from validated model-extracted brand-language observations, which remain model output. Parse's own domain appears among ChatGPT Search's ten most-cited websites in this window (a self-referential artifact of tracking); it stays in the aggregates, where it is 0.48% of citations, and is excluded from the named lists.

We deliberately left out run-to-run volatility, which has its own study (which AI engine changes its answers most), and older engine generations, which do not expose recommendation ranking and would not be comparable. All figures are measured in the Parse index over the stated window; they describe what the engines exposed, not why.

The report card says the two engines need two different plans. On Google AI Mode, citations are near-universal — 99.5% of answers carry them, each answer spreads across 11.5 websites, and 84.6% of citations go to sources a brand does not own, concentrated on YouTube, Google properties, Reddit, Medium, and LinkedIn. On that engine, the citation channel runs through large third-party platforms. On ChatGPT Search, your own website is the larger citation channel — 47.9% of citations are owned — but a sixth of answers cite nothing, and those answers offer no citation channel at all.

Perception work also splits: ChatGPT Search hedges 10.2% of recommendations and advises against brands more often; monitor it for reservations and negatives. Google AI Mode endorses at 77.9% positive; monitor it for whether you are in the answer at all. The scorecard is designed to be re-run quarterly on the same panel definition, so each row can be tracked as engines change.