Glossary · Beginner
What Is Temperature? Definition and Example
Temperature is a setting that controls how much randomness a language model uses when choosing each next token. Lower values give more predictable output, and higher values give more varied output. Google's documentation says lower temperatures "are good for prompts that require a less open-ended or creative response, while higher temperatures can lead to more diverse or creative results" (Google).
On this page
- How does temperature work?
- What's an example?
- Why does it matter for AI visibility measurement?
- What are the common misconceptions?
- From MentionX data: variation across engines
- Worked example: Ledgerline's one-off test
- How is temperature different from Top-P and Top-K?
- What should marketers do with this?
- Common questions
- Related
- Sources
Checked on 31 October 2026. "Our view" marks MentionX opinion. Ledgerline is a fictional company used for illustration.
How does temperature work?
At each step, the model assigns probabilities to possible next tokens. Temperature changes how it samples from them. Google explains that "a temperature of 0 means that the highest probability tokens are always selected". Even then, responses are "mostly deterministic, but a small amount of variation is still possible" (Google). Related settings include Top-P, Top-K and a seed (Google). If a model's response is too generic or too short, Google's guidance suggests trying a higher temperature (Google).
What's an example?
Fictional example: asked "Name a project management tool", a model at low temperature will usually give its single most likely answer. At a higher temperature it may name a different tool each time.
Why does it matter for AI visibility measurement?
Consumer AI search products don't publish their temperature or sampling settings, and answers vary from run to run. Our view: that's why a single test proves little. Run each prompt several times, report how often you're mentioned rather than a yes or no, and keep the number of runs the same between measurements. See mention rate and prompt set.
What are the common misconceptions?
- "High temperature means more accurate." It means more varied, not more correct.
- "Temperature 0 guarantees the same answer." Google says small variation is still possible (Google).
- "Temperature explains all variance." Retrieval results and model updates also change answers.
From MentionX data: variation across engines
Method. Read-only aggregate queries over MentionX scans from 22 August to 29 October 2026 (UTC): 2,217 engine runs (2,154 completed) across 18 anonymised projects and six engines.
Run-to-run variation is the practical face of sampling settings like temperature, along with retrieval and model updates. Across engines in our sample, brand mentions and citations differed widely:
| Engine | Brand mentioned | Answers with no citations |
|---|---|---|
| Gemini | 47.3% | 50.7% |
| OpenAI | 42.9% | 30.2% |
| Perplexity | 31.5% | 4.1% |
| AI Mode | 31.3% | 12.6% |
| Copilot | 26.0% | 47.0% |
| DeepSeek | 24.0% | 99.0% |
The question type matters as much as the engine:
| Intent | Brand mentioned | Share of voice |
|---|---|---|
| Brand-specific | 100% | 64.3 |
| Comparison | 91.9% | 44.6 |
| Recommendation | 29.2% | 26.8 |
| Best-of | 26.6% | 22.7 |
| Problem | 5.7% | 7.7 |
Our view: if a single test showed your brand missing, you know very little. Repeat each prompt, report a rate and compare like with like.
Limits. A convenience sample of mostly B2B software projects, run through engine APIs, which can behave differently from the consumer apps. Perplexity, Copilot and DeepSeek figures come from a single day (26 October 2026), so treat them as a snapshot, not a trend.
Worked example: Ledgerline's one-off test
Ledgerline is a fictional accounts-payable tool. Figures are illustrative.
Ledgerline's CEO asks an AI assistant, "What's the best accounts-payable tool for a mid-sized company?" Ledgerline isn't named, and the CEO asks the marketing team why. The team runs the same prompt ten times over two days on two engines. Ledgerline appears in some runs and not others, and the competitors named also change.
The team reports a mention rate rather than a yes or no, and sets up regular tracking with the same prompts and number of runs, so future changes can be compared fairly.
How is temperature different from Top-P and Top-K?
All three shape how a model picks the next token. Temperature changes how sharply the model favours high-probability tokens. Top-K limits the choice to the K most likely tokens, and Top-P limits it to the smallest group whose combined probability reaches a threshold. Developers can set these through APIs; users of consumer products usually can't.
What should marketers do with this?
Accept that variation is normal. Build a fixed prompt set, run each prompt several times on each engine, record the results consistently and report rates. If you use APIs to monitor answers, note your temperature setting, because it won't match the consumer product exactly.
Common questions
Can I set the temperature in ChatGPT? Not in the consumer app. API users can set it for their own requests.
Should monitoring use temperature 0? It reduces variation in API tests, but consumer products won't behave the same way, so record what you used.
Related
Sources
- Google Cloud, Experiment with parameter values (accessed 31 Oct 2026)
Questions
Does temperature 0 make answers identical?
Mostly, but not fully. Google says responses at 0 are mostly deterministic, but a small amount of variation is still possible.
Can I see the temperature ChatGPT or AI Overviews use?
No. Consumer AI products don't publish their settings, which is one reason the same prompt gives different answers.