Key Takeaways
- Treat a prompt as the whole request, not just the words typed. The assistant also reads its own instructions, the earlier conversation and sometimes the person's location or saved preferences.
- Expect every added detail to narrow the field. A city, a type of client, a goal or a budget each removes businesses that do not state that detail on a page software can read.
- Write down the details your buyers actually add. You cannot control the wording of a prompt. You can make sure your pages answer the details people put in it.
- Test one question in several wordings. Five versions of the same question tell you far more than one. Being named for the short version and dropped for the specific one is a finding.
- Do not chase every phrasing. Synonyms rarely change who is named. Details do. Spend your effort on the details, not on stuffing a page with every way to say "coach".
Every answer an AI assistant gives starts with a prompt. Change the prompt by a few words and the list of businesses in the answer can change completely.
So what is a prompt in AI, exactly? It is the request the assistant answers: the words a person types, plus a layer of things they never see. And the reason it matters to a coach or consultant is simple. Every detail in a prompt is a filter, and a business either passes it or does not.
Below is the short answer, what else travels with the typed words, how one question narrows as details are added, which wording changes matter and which do not, and what to do about it. The worked example is made up and labeled so.
What is a prompt in AI?
A prompt is the request an AI assistant answers. In everyday use it means the question or instruction someone types into ChatGPT, Claude, Gemini or Perplexity: "who is a good business coach in Denver?" is a prompt.
The word is also used more widely, for everything the model is given before it writes. Anthropic's glossary of AI terms describes the context window as the model's "working memory", and says a larger one lets it handle "more complex and lengthy prompts". In other words, the prompt is the whole working text the model reads, not only the line a person typed.
For a business, the useful definition is the narrow one plus a warning. The typed question is what you can learn from, and the rest is a reason two people asking the same thing can still see different answers.
What else is in the prompt besides the typed words
Besides the typed words, a prompt usually carries the assistant's own instructions, the earlier messages in the conversation, and sometimes the person's rough location or saved preferences. None of it is visible to you, and all of it can shape who gets named.
- The assistant's own instructions. Every consumer assistant runs with instructions from the company that makes it, written before the person typed anything.
- The conversation so far. If someone spent ten minutes describing their restaurant before asking for a coach, all of that is part of the request.
- Location and memory. Some assistants use an approximate location or remembered details from earlier chats, depending on the product and the person's settings.
- Search results. When an assistant searches the web, what it finds is added to the request before it writes the answer.
That last point changes what one prompt is. Google's page on AI features in Search says AI Overviews and AI Mode may use a "query fan-out" technique, "issuing multiple related searches across subtopics and data sources". One typed sentence can become several searches, and each detail in it can become a search of its own.
Why the same question asked twice can still come back different, even with identical wording, is a separate topic, covered in why the same question gives different answers. This post holds the question steady and changes the words.
How each added detail changes who gets named
Each detail added to a prompt removes the businesses that do not state it. The figure takes one made-up question and adds a detail per row, the way a buyer often refines what they ask.
-
Who is a good business coach?
Almost anyone with "business coach" on a page, plus the best-known names
-
Who is a good business coach in Denver
Coaches who say plainly that they work in or around Denver
-
Who is a good business coach in Denver for restaurant owners
Coaches whose pages name restaurants or hospitality clients
-
Who is a good business coach in Denver for restaurant owners who want to open a second location
Coaches who describe growth or expansion work, not only fixing problems
-
Who is a good business coach in Denver for restaurant owners who want to open a second location, and roughly what does it cost?
Coaches who publish a price or a starting fee the assistant can repeat
The question never changed. Each detail is a test, and a business passes it only if a page says so in plain words.
The pattern is the point, not the example. A coach who is named for the first row can vanish by the third, because nothing on their site mentions restaurants. A smaller coach who writes "I work with independent restaurant owners in Denver" can appear from the third row on, without being named at all for the short version.
The last row is the one most businesses fail. Buyers ask about cost all the time, and an assistant can only repeat a price that is written down somewhere it can read. If you publish none, the answer either guesses or names someone who did.
Which wording changes matter and which do not
Details change who is named. Synonyms mostly do not. "Coach" and "mentor", or "good" and "great", tend to produce similar answers, while adding a place, a client type, a goal, a format or a budget can replace most of the list.
The table sorts common wording changes by how much they tend to move an answer. The sorting is our reading from asking these questions, not a measured rule, and every example prompt is made up.
| Change to the prompt | Example | Tends to move the answer | What your page needs to say |
|---|---|---|---|
| Add a place | "...in Denver" | A lot | Where you work, or that you work online and with whom |
| Add a type of client | "...for restaurant owners" | A lot | Who your clients are, in their own words |
| Add a goal or situation | "...who want to open a second location" | A lot | The problems you solve and the results you work toward |
| Add a price condition | "...under $300 a month" | A lot | A price or a starting fee, written as text |
| Ask for one name instead of a list | "Who is the best..." | Some | A clear reason you are a fit, stated more than once and in more than one place |
| Swap a synonym | "mentor" for "coach" | Little | Your main job title, used consistently |
| Change the tone | "Please could you suggest..." | Little | Nothing extra |
That bottom half of the table is where a lot of wasted effort goes. Pages stuffed with every synonym for "coach" read badly to people and do little for the answers. One clear job title, used the same way everywhere, is enough.
The top half is where the work is. If the details your buyers add are not on your pages, no amount of rewording on your side will put you into the answers that use them.
What to do about it as a business
Write down the details your buyers add to their questions, make sure each one is stated on your site, and test one question in several wordings. You do not need to learn to write prompts. You need your pages to answer the prompts other people write.
Three steps, in order:
- List the details. Think about your last ten enquiries. What did people mention? A city, a stage of business, a deadline, a budget? Those are the details buyers will also type into an assistant.
- Check each one against your site. For every detail, find the sentence on your pages that states it. If you cannot point to the sentence, an assistant cannot either.
- Test a ladder. Take one question and ask it at three or four levels of detail, like the figure. Note the row where you appear and the row where you drop out.
For step three, the wording should sound like a client, not like a marketer. How to pick and word a full set of test questions is in how to pick the 25 questions that matter, and a ready list to start from is in questions to ask ChatGPT about your own business.
One honest limit. Nobody outside the AI companies knows which wordings people use most, or how often a given question is asked. The ladder tells you where you fit. It cannot tell you how many buyers stand on each row.
The free check asks ChatGPT 25 real buyer questions from your niche at $0 and emails a plain-English readout: your score, who got named in your place, and the one fix to start with. It runs as soon as you ask for it, and the report lands within the hour between 7am and 5pm Central, or by 10am the next morning outside those hours.
Questions we hear the most
What is a prompt in AI?
A prompt is the request an AI assistant answers: the question or instruction a person types, plus anything else the assistant is given alongside it, such as its own instructions and the earlier messages in the conversation.
Why does the wording of a prompt change which businesses AI names?
Because each detail in the question is something the answer has to satisfy. Adding a city, a type of client or a budget removes businesses whose pages do not state that detail, so a different set of names fits.
Is a prompt the same as a search query?
Not quite. A search query is usually a few keywords. A prompt is often a full sentence with a situation, a goal and conditions, and an assistant may turn one prompt into several searches before it answers.
Can a business control what prompts people type?
No. Nobody outside the AI companies sees what people type. What a business controls is whether its own pages state the details buyers tend to add, such as who it serves, where, in what format and at what price.
Do I need to learn prompt engineering to show up in AI answers?
No. Prompt engineering is the skill of writing requests to get better output from a model. Showing up is about what your pages say, not about how you write prompts. You only need to ask a few plainly worded test questions.
How does the free check choose which prompts to ask ChatGPT?
The free check asks ChatGPT 25 real buyer questions written for your niche, worded the way a client would type them, at $0. It emails your score, who got named in your place and the one fix to start with.
Does the $500 audit test more than one wording of a question?
The audit runs 125 checks across ChatGPT, Claude, Gemini, Perplexity and Google AI Overviews and keeps every answer so you can read it, landing by this time tomorrow. The free check covers ChatGPT only, with 25 questions.
Your buyers are already asking. Find out what AI tells them.
The free check asks ChatGPT 25 real buyer questions about your niche and sends you the report within the hour, at $0. If it shows you are already getting named everywhere, we will say so in plain words.
Get my free check →