AI model selection
Choose which AI models each prompt runs on
Track specific AI models where they matter. When you add prompts in Rankry, pick the engines they should run on, from ChatGPT and Perplexity to Google AI Mode. Leave the choice empty and the prompt runs on every engine in your plan.
7-day Base trial · No card needed
Add prompts · Models
4 prompts- ChatGPTSelected
- PerplexitySelected
- Google AI ModeSelected
- ClaudeOff
- GeminiOff
Sample data
12 requests for the first run (4 × 3 models) · 2,140 available
In short
AI model selection is choosing which AI engines a prompt is asked on. In Rankry, a prompt runs on every engine in your plan unless you pick a subset when you add it, in the “Add prompts” window or with a models column in a CSV import. Each prompt costs one request per selected engine per run, so a focused pick saves requests and keeps each engine’s numbers clean.
Nine AI engines
Which AI models can a prompt run on?
Rankry tracks nine AI engines. Which chips you can pick depends on your plan; engines your plan doesn’t include show as “Not on your plan”. The last column is our suggestion, not a rule.
| Engine | How it is asked | Worth including when |
|---|---|---|
| ChatGPT | OpenAI’s API, with web search | Almost always. Few categories can afford to leave it out. |
| Claude | Anthropic’s API, with web search | Your buyers are professionals, developers or teams who work in Claude. |
| Gemini | Google’s Gemini API, grounded in Google Search | Your buyers live in Google’s world: Android, Chrome, Workspace. |
| Perplexity | Perplexity’s API, with its search results | Your category is research-heavy and buyers compare options before deciding. |
| Grok | xAI’s API, with web search | Your audience is active on X, or news and opinion shape your category. |
| Meta Muse | Meta’s official API, with web search | You want Meta’s assistant next to the other chat engines on your core prompts. |
| Google AI Overviews | Captured from Google’s live results | The prompt reads like a search people type into Google. |
| Google AI Mode | Captured from Google’s live AI Mode results | Buyers ask longer, conversational questions inside Google. |
| Copilot | Azure GPT with Bing web search, through Microsoft’s official API | Your buyers search with Bing, or you want an answer grounded in Bing next to Google’s. |
Google AI Overviews and AI Mode are included on Pro or available as Base add-ons, and not for projects in Belarus, Iran or Russia. Claude, Grok, Meta Muse and Copilot are add-ons on Base and Pro.
Why engines differ
Why track more than one AI engine?
Each engine builds its answers in its own way. They run different searches, trust different sources and write lists of different lengths. A brand that ChatGPT names first can be missing from Perplexity, and an AI Overview can cite pages that no chat engine uses.
That is why Rankry reports every engine separately as well as together. The Model comparison table on the dashboard shows Visibility, Position, Sentiment and Diversity for each engine side by side, with the best cell highlighted, and Diversity counts how many engines name you at all.
Choosing engines per prompt lets you spend requests where those differences matter. Run the questions that define your category everywhere, and give narrower prompts only the engines your buyers use for them.
In the product
How do I choose AI models for a prompt?
You set engines when you add prompts. It takes a few seconds, and the cost is shown before you confirm.
- 01
Open “Add prompts”
On the Prompts page, open the Add prompts window. Write one prompt per line, or switch to “Import CSV”.
- 02
Pick a category
File the prompts under an existing category or create a new one. The default is “Custom Prompts”.
- 03
Select the models
Click the engine chips you want. Leave them all on to run on every engine in your plan.
- 04
Check the cost line
The window shows the requests for the first run, prompts × models, and how many requests you have available.
- 05
Add the prompts
They start answering right away on the engines you picked, and from then on they run with every scheduled run.
- 06
Check the Models column
In the prompts table, logos show which engines each prompt runs on.
Other ways to set engines
Can I set AI models in a CSV import or during onboarding?
Yes to both, with a few rules worth knowing before you build a large prompt list.
CSV import on the Prompts page
Upload a .csv or .txt file with a header row. The “prompt” column is required; add an optional “models” column listing engine names, separated by commas.
Onboarding, on plans that include it
When you review your first prompts, the “AI models for these prompts” picker lets you choose engines for your starting set. On other plans this step shows “All models”.
Empty means everything
A prompt with no selection follows your plan. It runs on every engine the plan includes at the time of each run, including engines your plan gains later.
Fixed once added
After a prompt is added, its engines can’t be changed. Its history therefore always compares the same engines, run after run.
Fixed engines
Why can’t I change a prompt’s models later?
In Rankry, a prompt’s engines are set once, when you add it, and there is no way to change them afterwards. That has a useful side effect: every run in a prompt’s history answers the same question on the same engines, so a change in the trend reflects a change in the answers, not a change in what was measured.
If you need a different set of engines for the same topic, add a new prompt worded for that purpose with the engines you want. Its history starts with its first answers, which arrive right away.
Prompt text works the same way. You can edit wording while you review prompts at onboarding; after that, prompts are added and removed rather than edited. A little planning up front saves requests later.
Cost and metrics
How does model selection affect cost and metrics?
Picking fewer engines changes two things: how many requests a prompt uses, and how much it adds to each engine’s numbers.
| Area | Effect |
|---|---|
| Requests per run | Prompts × selected engines. 20 prompts on 3 engines use 60 requests a run; on all 9 engines, 180. |
| First run | Charged when you add the prompts, and shown in the cost line before you confirm. |
| Visibility | The share of prompt-and-engine answers that name you. A prompt on fewer engines adds fewer answers to the total. |
| Diversity | How many of the engines that ran named you. |
| Model comparison | Visibility, Position, Sentiment and Diversity for each engine, from the prompts that ran on it. |
| Compete | The per-engine views on the leaderboard, share of voice, gap to leader and radar use the answers each engine gave. |
Failed answers and sentiment scoring don’t use requests.
Practical mixes
How should I pick engines for different kinds of prompts?
There is no single right mix. These patterns work well for most brands and keep request use in check.
Core buying questions
Run them on every engine. These prompts define your visibility, and the differences between engines are often the most useful finding.
Google-style searches
Short, keyword-like prompts fit Google AI Overviews and AI Mode, where people type searches.
Long research questions
Conversational, multi-part questions suit ChatGPT, Claude, Perplexity and AI Mode.
Local questions
Prompts with a city or neighborhood are worth running on Google’s surfaces, where local searches already happen, plus the chat engines your customers use.
Testing a hunch
Want one engine’s view of a niche topic? Add the prompt on that engine alone and save requests.
A tight budget
Keep your core set on all engines and run the long tail on two or three.
Explore Rankry
Related features
Can I choose which AI models to track?
Yes, prompt by prompt. When you add prompts in the “Add prompts” window, click the engine chips you want, or list engines in a “models” column of a CSV import. On plans that include it, you can also choose engines for your starting prompts during onboarding.
What happens if I don’t select any models?
The prompt runs on every engine your plan includes, at every run. If your plan adds an engine later, prompts without a selection pick it up automatically.
Can I change the models for a prompt after adding it?
No. Engines are fixed when a prompt is added, which keeps its history consistent. To track the same topic on other engines, add a new prompt with the engines you want; it is answered right away.
Does selecting fewer models save requests?
Yes. Each prompt uses one request per selected engine per run. Ten prompts on two engines use 20 requests a run instead of 90 on all nine. The Add prompts window shows the cost of the first run before you confirm.
Why can’t I select some engines?
Engines your plan doesn’t include show as “Not on your plan” and can’t be selected. On the dashboard, engines outside your plan appear grayed out or blurred until you move to a plan that includes them.
Are Google AI Overviews and AI Mode included?
On Pro, yes: they appear as chips like any other engine. They are captured from Google’s live results rather than asked through an API, and they are not available for projects in Belarus, Iran or Russia.
Can I pick models during onboarding?
On plans with model selection, the prompt review step has an “AI models for these prompts” picker. On other plans the step shows “All models” and your starting prompts run on every engine in the plan. Prompts you add later can carry their own engines on any plan.
How do I set models in a CSV file?
Use a header row with a required “prompt” column and an optional “models” column that lists engine names separated by commas. Leave the models cell empty for a prompt to run on every engine in your plan. Each prompt needs at least 3 characters and can be up to 200.
Should I track every prompt on every engine?
For your core buying questions, usually yes: differences between engines are often the most useful insight. For long-tail or exploratory prompts, a smaller set saves requests without losing much.
Track the AI engines your buyers actually use
Pick engines prompt by prompt and see your first answers right away.
7-day Base trial · No card needed