Glossary
What is a prompt?
A prompt is the question or instruction sent to an AI assistant. In AI search analytics, it is the starting point of every record you keep.
What a prompt is
A prompt is whatever you send to an AI assistant to get a response. It can be a question, such as which accountants near me handle sole traders. It can also be an instruction, such as asking for a shortlist of local florists.
In a chat app, the prompt is what a person types. Behind the scenes, the service may add its own instructions too. OpenAI's guide explains that developer instructions take priority over the user's prompt, so the answer depends on more than the words typed.
Prompts in AI search analytics
In AI search analytics, a prompt is a question a business chooses to put to AI assistants on purpose. The aim is to see what those assistants say when a customer asks something similar.
A prompt that names your business is a branded prompt. One that describes a need without naming you is non-branded. Keep the two apart, because a prompt that names you will often get an answer that names you too.
Why the same prompt can give different answers
OpenAI's guide says the content a model generates is non-deterministic. In plain words, the same prompt can produce a different response each time it is sent.
Google's help pages for AI Overviews suggest asking several versions of a question to get the best answers. That is advice to searchers, and it also shows how much wording matters.
So a single answer is one observation. It tells you what happened on that occasion. Patterns only show once the same prompt has been asked several times under the same conditions. Anthropic's guide makes a related point for builders, treating a prompt as something to test against agreed criteria. If an attempt produced no readable answer, record that too. A missing answer does not show that your business was left out. It shows only that nothing could be read that time.
How Carqo relates to prompts
Carqo's Prompts area is built to hold the list of questions tracked for your website. Each saved prompt sits in one library beside its recent results, and links through to the dashboard for its set.
Each prompt is saved as a revision. Answers collected under earlier wording stay tied to that wording, so a change to a prompt never rewrites its history.
In practice
- Collect real customer questions Gather questions from enquiries, calls and reviews. Note where each one came from.
- Write them in the customer's words Keep the phrasing a customer would use, including place names. Save the exact wording before you send it.
- Record the context Note the date, assistant, model, country and language for each attempt. Keep failed attempts as well as answers.
Common questions
Is a prompt the same as a search query?
They are close relatives. A search query is often a few words, while a prompt can be a full question or instruction.
Why does an assistant answer the same prompt differently?
OpenAI says model output is non-deterministic, so the same prompt can produce different responses. Model updates can change answers too. Treat each answer as one observation and look for patterns across repeated checks.
How many prompts should a business check?
There is no correct number. Start with the questions tied to your main services and the places you serve. Add more only when you can review the answers properly.
Is a prompt the same as a search query?
They are close relatives. A search query is often a few words, while a prompt can be a full question or instruction.
Why does an assistant answer the same prompt differently?
OpenAI says model output is non-deterministic, so the same prompt can produce different responses. Model updates can change answers too. Treat each answer as one observation and look for patterns across repeated checks.
How many prompts should a business check?
There is no correct number. Start with the questions tied to your main services and the places you serve. Add more only when you can review the answers properly.
Sources
- OpenAI API documentation, Prompt engineering Link checked 29 October 2026
- Anthropic documentation, Prompt engineering overview Link checked 29 October 2026
- Google Search Help, AI Overviews in Google Search Link checked 29 October 2026