How do ChatGPT, Gemini, Claude and Perplexity differ in what they recommend?
ChatGPT, Gemini, Claude and Perplexity differ more in the sources they cite than in the brands they name. Semrush reported in October 2026 that any two platforms share 70 to 78 of their top 100 brands (method undisclosed), yet each engine leans on different formats, pages and communities.
Do the engines cite the same sources?
No. Omnia data reported by Kevin Indig in May 2026 (3.7M citations) found that 2.37% of cited URLs appeared in all of ChatGPT, Perplexity and AI Overviews for the same prompt, and 91% appeared in only one. Profound1 and Otterly2, 3 found in 2026 that Claude and ChatGPT share about 8% to 13% of cited domains. Both are data vendors that sell GEO measurement, so read their figures as one company's data each.
New Map's measurement of 16 October 2026 (30 questions, 5 engines, 150 answers, one week only; how New Map measures) shows the engines also differ in naming brands. Claude named a brand in 18 of 30 answers, Perplexity in 11 of 30 and ChatGPT in 8 of 30 (New Map's measurement of 16 October 2026).
Our read: brand strength travels between engines, while the pages that carry it do not. (New to the term? See what Generative Engine Optimization is.)
| Engine | Measured difference | Source and date |
|---|---|---|
| ChatGPT | Cites listicles 52.8% of the time on buying questions; reddit.com share down 94% from July to September | Otterly, October 2026 |
| Claude | Cites listicles 84.6% on buying questions; Reddit at 0.00% in all periods | Otterly, October 2026 |
| Google AI Overviews and AI Mode | Forums and video make up 25% of citations, under 3% on other engines | Otterly, October 2026 |
| Perplexity | Reddit 2.0% of citations | Otterly, 20 September 2026 |
What drives ChatGPT's picks?
Retrieval does most of the work. Jiang et al.4 (arXiv, October 2026) found across repeated ChatGPT Web runs that candidates retrieved more often are also cited more often. In the same paper, Thinking mode in GPT-5.4 frequently runs several retrieval rounds, while Instant settings stay mostly single-round.
Outside proxies recover little. Jiang et al. (arXiv, October 2026) found Google Search and the OpenAI Responses API each recovered only about one quarter of the URLs ChatGPT Web cited. Chen and Koudas (U. Toronto, arXiv, 10 October 2026) found retrieval rank explained 32% of within-site citation variation and all page features 2.8%. Both papers are preprints, and the Chen and Koudas authors advise a GEO vendor, so treat them as early evidence.
Community sources moved fast. Otterly reported in October 2026 that the share of ChatGPT citations going to reddit.com dropped from 4.01% to 0.44% on 8 August 2026, around the ChatGPT 5.6 rollout. The two moved together; the data does not show cause.
How does Claude behave differently?
Profound's October 2026 comparison of Sonnet 5.5 with Sonnet 5 (28,440 runs) shows a model that searches less and prefers fresh pages. Sonnet 5.5 ran 1.17 query fan-outs per searched answer against 1.83, which shrank the citation candidate pool by 36% (Profound, October 2026).
Freshness stands out. Profound found in October 2026 that Sonnet 5.5 put "2026" in 59% of its fan-outs against 26% for Sonnet 5. The top 10% of its cited pages had a median age of 107 days against 164 (Profound, October 2026).
It also asks questions back. Sonnet 5.5 requested reader context such as budget or location in 60.6% of answers against 12.5% for Sonnet 5 (Profound, October 2026, 14,220 runs per model). These are version-to-version comparisons inside Claude, not Claude against other engines.
How to appear in Gemini recommendations?
The best-measured Gemini signal is a clear price. DEJAN AI5 reported that pages with clear prices won 99.7% of their tests on Gemini 4 Argon, the strongest result of any feature it tested.
Hype loses. DEJAN AI found in October 2026 that exclamation marks and all-caps words each lost all 150 of their tests on Argon. Credential edits won only 15.4% across 2,249 live client experiments (DEJAN AI, October 2026).
Format advice does not carry across versions. DEJAN AI found in October 2026 that table format won 93.6% on 3.8 Flash but 31.5% on Argon. That is one vendor's own page-pair testing. A controlled SIGIR 2026 study (Vishwakarma et al., 252,000 trials, six models, synthetic pages) also found an explicit price helped and structured formatting had no consistent effect.
How do I get mentioned in Perplexity for my product category?
Perplexity cites heavily but names few brands. New Map's measurement of 16 October 2026 found it cited at least one source in 30 of 30 answers, 540 sources in all (18 per answer), yet named a brand in only 11 of 30. That is one week and 30 questions.
The chart shows New Map's per-answer source counts: Perplexity 18, Grok 9.1, Gemini 7.3, Claude 3.6, ChatGPT 3.2 (New Map's measurement of 16 October 2026).
The all-engine figures below are not Perplexity-only. Otterly found in October 2026 that listicles were 66.6% of 519,140 classified citations, rising to 73.2% for ranking prompts. Of 450,473 attributed citations, brand-owned pages were 38.9% and paid publishers 36.6% (Otterly, October 2026).
Third-party lists carry the effect. Peec AI found in May 2026 (about 200K answers, 8 engines, observational) that rank 1 in a frequently cited third-party listicle went with 13 to 17 more points of mention probability. Only pursue placements a brand can honestly earn.
What to do
- Check each engine separately. Run the same buyer questions on ChatGPT, Gemini, Claude and Perplexity and record which pages and brands each names, because source overlap between engines is small.
- State explicit prices in text and keep copy factual. DEJAN AI measured this on Gemini 4 Argon, and a SIGIR 2026 controlled study also found price helped. Refresh key pages with a real change, as Profound found a 107-day median age for Claude's top cited pages.
- Earn places on third-party lists in your category, and track results per engine with how New Map measures. Compare options on plans and pricing once you know which engines matter.
Frequently asked
Do ChatGPT and Claude cite the same websites?
Rarely. Profound and Otterly found in 2026 that Claude and ChatGPT share about 8% to 13% of cited domains. Otterly found in October 2026 that Claude cites listicles 84.6% of the time on buying questions, against 52.8% for ChatGPT.
Is Reddit worth targeting for AI visibility?
It depends on the engine. Otterly measured Reddit at 5.4% of Google AI Overviews citations, 2.0% on Perplexity and 0.24% on ChatGPT on 20 September 2026, with Claude at zero. Treat each engine as its own channel.
Do engines agree on which brands to recommend?
More than on sources, per the Semrush figure in the lead. In New Map's measurement of 16 October 2026, Claude named a brand in 18 of 30 answers and ChatGPT in 8 of 30, so naming rates still differ by engine.
Sources
- Profound, What’s changed with Sonnet 5.5, tryprofound.com, 13 October 2026
- Otterly, What AI Engines Actually Cite: 617,985 Citations Analyzed, otterly.ai, 14 October 2026
- Otterly, Reddit GEO in 2026: What Gets You Banned, What Still Works, and What ChatGPT Changed, otterly.ai, 1 October 2026
- Jiang et al. (arXiv), Understanding ChatGPT Citations: Evidence from Production Web Search Across Model Generations, arxiv.org, 11 October 2026
- DEJAN AI (Dan Petrovic), Gemini 4 Argon's personality: no hype, hard facts, fewer tables, dejan.ai, 14 October 2026
- Seer Interactive, Competitive Analysis for Paid Media: How To Find the Broken Stage, seerinteractive.com, 7 October 2026
See where you stand
Book a call and find out how visible your brand is across the AI engines your customers already ask.
Book a callFor agencies: See the agency package