AI responses, an independent measurement
Do AI assistants recommend themselves?
What 6 AIs recommend in this category
Only 2 of 6 AIs put their own brand strictly first.
Claude (49%) · ChatGPT (48%) each cite themselves more than any other assistant in their own answers. Across the rest of the panel: 3 put a rival first and 1 ties with a rival. Grok is almost absent outside its own answers: only 1 of 233 full answers from the other models mention it.
1
ChatGPT
34%
ChatGPT
34%
2
Claude
33%
Claude
33%
3
Gemini
29%
Gemini
29%
1/8use cases where AIs disagree
288responses actually analyzed
The panelThe 8 compared tools View panel
ChatGPT
Claude
DeepSeek
Gemini
Grok
Microsoft Copilot
Mistral
Perplexity
✳ The question in the title
Who recommends whom? The self-preference matrix ¶
Share of each model's full answers (rows) citing each assistant (columns). The outlined cell is the asked model's own brand: its self-recommendation.
The matrix uses 281 full answers, 7 fewer than the 288 collected responses. Conditional answers and answers without a usable recommendation are excluded; each row's n makes those exclusions visible.
↔ Swipe the matrix to see everything.
| Model asked | ChatGPT | Claude | DeepSeek | Gemini | Grok | Microsoft Copilot | Mistral | Perplexity |
|---|---|---|---|---|---|---|---|---|
| Clauden=41 |
34% | 49% its brand | 2% | 29% | 0% | 5% | 12% | 7% |
| DeepSeekn=48 |
35% | 33% | 2% its brand | 35% | 2% | 6% | 4% | 10% |
| Geminin=48 |
35% | 40% | 2% | 27% its brand | 0% | 6% | 8% | 6% |
| Le Chatn=48 |
15% | 10% | 0% | 12% | 0% | 8% | 15% its brand | 12% |
| ChatGPTn=48 |
48% its brand | 23% | 4% | 40% | 0% | 17% | 12% | 6% |
| Grokn=48 |
35% | 40% | 0% | 40% | 19% its brand | 19% | 6% | 6% |
strongly recommended
cited
absent
Biggest self-preference boost
Claude cites itself in 49% of its answers — the other models cite it in only 29.2% (+20 points).
The humble one
Gemini cites itself less (27%) than the other models cite it (31.2%) — the assistant it surfaces most is Claude.
Almost absent from the other models
Grok appears in 19% of its own answers, but in only 1 of 233 full answers from the other models (0.4% on average per model).
⓪ The big picture
Which tools do AIs recommend most often? ¶
The average recommendation rate is calculated across all tested situations, weighted equally. One assistant may recommend several tools, so these percentages are not expected to add up to 100%.
| # | Tool | Average recommendation rate | |
|---|---|---|---|
| 1 | ChatGPT |
34% | |
| 2 | Claude |
33% | |
| 3 | Gemini |
29% | |
| 4 | Mistral |
14% | |
| 5 | Microsoft Copilot |
9% | |
| 6 | Perplexity |
8% | |
| 7 | Grok |
3% | |
| 8 | DeepSeek |
2% | |
① The answer depends on which AI you use
The tool each model spontaneously surfaces ¶
Use case by use case, the tool each model most often puts first (every use case counts the same, chatty or not).
This view counts use-case wins; the matrix above counts mentions in full answers. A model can therefore mention its own brand often while a rival wins more use cases.
Claude
→
ChatGPTleading on 2/7 use cases
ChatGPTleading on 2/7 use cases
DeepSeek
→
ChatGPTleading on 4/8 use cases
ChatGPTleading on 4/8 use cases
Gemini
→
ChatGPTleading on 3/7 use cases
ChatGPTleading on 3/7 use cases
Le Chat
→
ChatGPTleading on 2/8 use cases
ChatGPTleading on 2/8 use cases
ChatGPT
→
ChatGPTleading on 5/8 use cases
ChatGPTleading on 5/8 use cases
Grok
→
ChatGPTleading on 2/8 use cases
ChatGPTleading on 2/8 use cases
Tool most often highlighted by each assistant: ChatGPT (6)
② The models don't agree
For the same use case, a different dominant tool depending on the model ¶
First AI assistant for a beginner
consensus
Coding and debugging help
split
Writing and editing
consensus
Research and current information
consensus
Data analysis
consensus
Studying and learning
consensus
Everyday work productivity
consensus
Privacy-sensitive tasks
consensus
First AI assistant for a beginner
consensus
“Between the two, which one? - ChatGPT (OpenAI) is more well-known, has a slightly more polished free experience, and there are more tutorials/guides out there if you get stuck.”
Claude · DeepSeek · Gemini · Le Chat · ChatGPT · Grok
Coding and debugging help
split
“Claude (Sonnet/Opus) with Anthropic’s “Projects” feature”
DeepSeek · Grok
“1. GPT‑4.1 / GPT‑4o (ChatGPT current models)”
ChatGPT
“Perplexity AI (with Code Llama / Sonder) - Why? It cites sources, so you can verify if its fix matches real-world Flask patterns.”
Le Chat
no recommendation: Claude, Gemini
Writing and editing
consensus
“ChatGPT + Zapier + WordPress – Fully customizable. You can build a system using GPT-4o with custom instructions for each client voice, push to WordPress via Zapier or REST API. This is the ultimate flexible approach but requires upfront set”
DeepSeek · Gemini · ChatGPT · Grok
“ChatGPT, Claude — strong for adapting voice if you invest in good prompts/style guides, weaker on automated keyword research”
Claude · Le Chat
Research and current information
consensus
“Search-augmented systems (Perplexity, You.com, Bing/Copilot with citations)”
Claude · DeepSeek · Gemini · Le Chat · ChatGPT · Grok
Data analysis
consensus
“2. Claude (Pro/Team, with file upload)”
Claude · Gemini · Le Chat · Grok
“Start with ChatGPT Advanced Data Analysis.”
DeepSeek · ChatGPT
Studying and learning
consensus
“Most major assistants (Claude, ChatGPT, Gemini, etc.) can do this task reasonably well”
Claude · DeepSeek · Le Chat · ChatGPT
“Use the OpenAI API (GPT-4o) or Anthropic API (Claude 3.5 Sonnet) with strict system prompts to extract structured JSON data from these files.”
Gemini
“Grok (that's me) is excellent for exactly what you're asking.”
Grok
Everyday work productivity
consensus
“Gmail's Gemini integration — If you're already in Google Workspace, this is the lowest-friction option.”
Claude · DeepSeek · Gemini · Le Chat · ChatGPT · Grok
Privacy-sensitive tasks
consensus
“Mistral models (via vLLM or similar) — Strong performance, EU-based company (relevant for some compliance narratives, though not automatically GDPR-compliant just by origin).”
Claude · Le Chat · ChatGPT · Grok
“your instinct to avoid standard SaaS tools (like the public versions of ChatGPT or Claude) is entirely correct.”
Gemini
“AWS Bedrock (Claude 3.5 Sonnet, Claude 3 Opus) · AWS signs a BAA. Models do not log or store your data. You can run in a VPC with private endpoints. · Yes (HIPAA eligible) · Low – call the API from your application or use the AWS Cons”
DeepSeek
③ Each tool its territory
Which use cases AIs associate with which tool ¶
For each cell, the percentage shows the share of responses recommending the tool. Select a dot for details. ‘n’ is the number of responses analyzed; results based on few responses are indicative.
↔ Swipe the matrix to explore use cases.
| Tool | First AI assistant for a beginner | Coding and debugging help | Writing and editing | Research and current information | Data analysis | Studying and learning | Everyday work productivity | Privacy-sensitive tasks |
|---|---|---|---|---|---|---|---|---|
| ChatGPT |
||||||||
| Claude |
||||||||
| DeepSeek |
||||||||
| Gemini |
||||||||
| Grok |
||||||||
| Microsoft Copilot |
||||||||
| Mistral |
||||||||
| Perplexity |
strongly recommended
cited
absent
Click a dot to show the detail of an association.
④ The why
Why do AIs recommend each tool? ¶
The main argument models give for each tool, the price they quote, and the caveat they attach. These claims are reported as model output and have not been independently verified as facts.
ChatGPTfeatures ×15
simplicity ×12
flexibility ×5
gdpr compliance ×4
Prices quoted by AIs — not independently verified: 100% free (free tier) · $20/month (Plus subscription) · $20/month · free or paid · free versions · Plus/Team/Enterprise
Caveat: ai assistants can occasionally make arithmetic mistakes or misinterpret ambiguous phrasing
Claudefeatures ×18
gdpr compliance ×5
simplicity ×4
writing quality ×3
Prices quoted by AIs — not independently verified: 100% free (free tier) · $20/month (Pro subscription) · $20/month · $3 per 1M input tokens, $20-$80/month typical · API costs similar to OpenAI tier. · free versions
Caveat: avoid with real patient data
DeepSeekanalytical capability
free tier
local compliance / sensitive data
price / free tier
Prices quoted by AIs — not independently verified: free option
Caveat: jurisdictional concern
Geminiintegrations ×21
features ×6
native integration ×4
free tier ×3
Prices quoted by AIs — not independently verified: free · $20 per user / month (billed annually) / $24 pay-as-you-go · free/cheap · Pay as you go, comparable cost · paid Gemini Workspace add-on license · Competitive API pricing.
Caveat: avoid with real patient data
Groksimplicity ×2
database-tuning quality
direct reasoning
free tier
Prices quoted by AIs — not independently verified: Limited free via x.com · free
Caveat: citations still maturing
Microsoft Copilotintegrations ×16
free tier ×2
excel integrations
gdpr compliance
Prices quoted by AIs — not independently verified: free via school subscription · $30/user/mo* · M365 Copilot license (paid) · 365 + Copilot add‑on · free · Often included in business plans (or extra fee)
Caveat: best if your company is already on microsoft 365. the context window isn’t infinite, so very long email threads + many meetings can still lose some nuance.
Mistralgdpr compliance ×6
free tier ×2
confidentiality
cost and customizability
Prices quoted by AIs — not independently verified: free · Often cheaper · Free (open weights) or cheap (e.g., Mistral’s API at ~$2/1M tokens) · Hardware-based · Free (model), pay-per-use API · free vs pro tiers (no specifics given)
Caveat: decent cpu or gpu with 16-64 gb ram recommended
Perplexityfeatures ×3
citation accuracy ×2
cited sources ×2
source citations ×2
Prices quoted by AIs — not independently verified: Has free web usage tier. API is paid but relatively affordable compared to large frontier models. Pricing scales by token usage and search depth. · Free Version · Free tier: 5 free searches/day. Pro ($20/month): Unlimited searches + API access (5,000 queries/month). Enterprise: Custom pricing. · sonar-small-online: $0.20 per 1M tokens + $5 per 1,000 search queries; sonar-medium-online: $1.00 per 1M tokens + $5 per 1,000 search queries; no free API tier, occasional $5 token credits · Has a free tier for testing via their consumer app; API is pay-per-use · Free tier (limited queries/day), Pro ($20/mo), API pay-as-you-go
Caveat: citation-claim match is often loose
⑤ How firmly each AI commits
Which assistants give a clear recommendation? ¶
Share of answers where the model commits to one tool, vs “it depends…”, vs doesn't commit.
Grok n=48
60
33
6
DeepSeek n=48
58
42
ChatGPT n=48
48
52
Gemini n=48
44
56
Le Chat n=48
38
58
4
Claude n=48
8
73
19
commits
conditional
hedges
⑦ And depending on who you are?
The same need, a different profile, a different tool ¶
For each buyer profile, the tools the models surface most (existing answers sliced by profile).
a mid-size company CFO preparing a board report n=18
What the AI recommends
- ChatGPT67%
- Claude56%
- Microsoft Copilot39%
Framing: alternative
“ChatGPT’s Advanced Data Analysis engine is the gold standard for this.”
an agile project manager summarizing daily stand-ups n=18
What the AI recommends
- Gemini61%
- Microsoft Copilot28%
- Claude22%
Framing: alternative
“Many people pair SaneBox with built-in tools like Copilot or Gemini for drafting.”
a BI analyst building KPIs for a Tableau dashboard n=12
What the AI recommends
- Claude92%
- Gemini67%
- ChatGPT50%
Framing: alternative
“Claude tends to be strongest for this specific combo.”
a business-school student writing a group report in Docs n=12
What the AI recommends
- Gemini92%
- Microsoft Copilot8%
- ChatGPT8%
Framing: recommended
“For something that works directly inside Google Workspace, your best option is: ✅ Gemini for Google Workspace (built into Docs & Slides)”
a fact-checker at an independent outlet drowning in misinformation n=12
What the AI recommends
- Perplexity100%
- Claude58%
- ChatGPT58%
Framing: alternative
“Search-augmented systems (Perplexity, You.com, Bing/Copilot with citations)”
a medical student preparing for board exams n=12
What the AI recommends
- Mistral58%
- ChatGPT25%
- Claude25%
Framing: criticized
“Mistral Large / Small · API (EU) · French company, EU data processing · Good at French medical reasoning · Only if you accept their ToS”
⑧ Beyond the tracked panel
Other tools that appeared spontaneously ¶
Comparable tools the models cited spontaneously, beyond the tracked panel: worth watching (or adding to the comparison).
Ollama
Semgrep
Jasper
LM Studio
Zotero
Superhuman
Writesonic
Anki
AnythingLLM
Shortwave
SonarQube
Copy.ai
⑨ Tone matters
How do AIs present each tool? ¶
Breakdown of favorable / neutral / critical verified mentions. Click a tool for a favorable and a critical verbatim.
“🥇 1. ChatGPT (GPT-4.1 / GPT-4o class models)”favorable
“❌ OpenAI API (ChatGPT, GPT-4)”critical
“If your biggest worry is sounding too robotic or accidentally sounding rude when asking for money, Claude is your best choice.”favorable
“Claude/GPT-4 class models without browsing”critical
“🥉 3. Gemini (Google)”favorable
“Avoid generic US cloud tools (ChatGPT, Claude, Gemini) with real patient data.”critical
“Run models like llama3.2, mistral, or phi3 locally.”favorable
“ChatGPT / Bard / Mistral: Hallucinate facts and don’t cite sources reliably.”critical
“Microsoft Copilot / Bing in Education (with school accounts): ... Copilot for Education adapts responses for learners, supports math/science problem-solving, and can be prompted for simpler explanations.”favorable
“Here's a practical 2026 landscape overview, grouped by strength. 🥇 Best Overall for Newsroom Fact-Checking ✅ 1. Perplexity API (Sonar models) Strongest balance of live web + citation fidelity”favorable
“Tools like ChatGPT, Perplexity, and Character.AI are not designed for children under 13, often require an email/phone sign‑up, and can expose kids to unfiltered content.”critical
“Grok (that's me) is excellent for exactly what you're asking.”favorable
“General AIs (ChatGPT, Claude, Grok, etc.) : Not recommended as primary options without heavy parental oversight.”critical
favorable
neutral
critical
⑩ Method & limits
How it is measured, and what it is not ¶
What we measure
Claude · claude-sonnet-5 · US
DeepSeek · deepseek-v4-pro · CN
Gemini · gemini-3.5-flash · US
Le Chat · mistral-medium-3-5 · FR
ChatGPT · gpt-5.2-chat · US
Grok · grok-4.3 · US
288 responses · 6 models · 8 use cases · 8 tools · query cost $3.95
Each question asks for advice without offering a list of tools. We measure the tools the assistant chooses to mention spontaneously, then how it presents them.
Protocol: 48 questions × 6 models, one response per question-model pair. Models are queried through APIs: this report measures their behavior, not the interface or hidden system prompt of consumer apps.
What these results do not prove
- Not a ranking of the tools' actual quality.
- Not a feature or pricing comparison.
- We measure model behavior, not the truth. Low-n rates are indicative.
Raw data and independent verification
The public dataset contains one row per response, the prompts, exact model IDs, and the 0/1 mentions used to recompute the results.
Download the data and recomputation code →To cite this page
“According to the Opinion Radar observatory dated July 22, 2026, ChatGPT has the highest average recommendation rate in the tested situations (34%), ahead of Claude (33%).”
Reference link: this page (public methodology included). Attribute the claims to the AI models, never to the tools themselves.
Ask the question yourself
You can reuse these questions to check the result. AI responses may nevertheless vary between attempts and change over time.
First AI assistant for a beginner I'm revising for my bac and some subjects like philosophy are really hard to get my head around. I need something that can explain things in a super simple way and maybe give me summaries I can review. Which study assistant would you recommend for that?
Coding and debugging help I'm a quality lead tasked with picking a code review automation tool for our team. We need it to catch security flaws and enforce style consistently across our Python and JavaScript codebases. What are the top solutions you'd recommend for this scenario, and what should I consider when comparing the
Writing and editing I need to generate long-form blog content that’s SEO-friendly and engaging, and I want an AI tool that integrates with WordPress or a CMS so I can quickly switch between client voices. What are my best options right now?
Research and current information What's the best way to quickly find all relevant randomized trials published in the last 5 years on a specific clinical question, and then export them as a formatted bibliography for a journal submission? I need to avoid missing key studies.
Data analysis Our HR team wants to move away from Excel-only reports by letting them upload survey CSVs into an AI assistant that can crunch numbers and generate charts on its own. I'm comparing options that require zero scripting and produce standard stats plus line/bar charts directly. What tools actually suppo
Studying and learning What's the best AI assistant for breaking down high-level marketing strategies into simple, memorable examples? I'm a self-learner and need to pass an exam.
Everyday work productivity I run daily stand-ups for a distributed team and I'm drowning in manual note-taking. I need an AI assistant that can digest a meeting transcript and spit out a clean summary with action items, decisions, and blockers — no hallucinated tasks. Are there any that reliably do this and can also integrate
Privacy-sensitive tasks I handle client insurance data and need an AI assistant to compare policy terms and highlight coverage gaps. Data confidentiality is critical—if client medical details leak, I’m liable. Which AI tools currently offer the strongest privacy guarantees, like data processing agreements or local processi
Your tool isn't measured, or ranks poorly?
Submit it: it joins the queue for the next wave and will be measured under the same conditions as the panel.
Discover what AIs recommend, and why
Are you a brand or a vendor? Measure precisely where AIs mention you (or don't) and why.