Before you pay for anything, ask the assistants yourself. It costs nothing and it teaches you a lot. This page shows how to do it properly, what it cannot tell you, and where a structured check starts to help.
The short answer
Asking by hand is a good first look. A structured check is better when you want a result you can repeat, compare and act on. Neither one can promise that an assistant will recommend you.
How to ask by hand, step by step
Most people ask one question once and trust the answer. A better routine takes about twenty minutes.
- Write down five things a buyer would ask when they do not know your name. For a fictional bakery in Leeds, that might be "Where can I order a wedding cake in Leeds?" or "Which bakeries near me do gluten-free birthday cakes?"
- Write each one in three different ways. Real people word the same need differently.
- Open a fresh chat for every question. A long chat carries memory of what you said before, and that bends the answer.
- Sign out of your accounts, or use a private window where you can. Saved memory and your location may colour the answer.
- Ask each question at least twice, on two different days if you can. Answers can change from one try to the next, even when nothing about the question changed. Our post on why AI answers change explains more.
- Turn on web search if the assistant lets you choose. Answers that use the web can differ from answers given from memory alone.
- Write down the result in a sheet, using the template below.
A simple tracking sheet
Copy these columns into any spreadsheet.
| Date | Assistant | Exact question | Were we named? | Position in the list | Who else was named | Sources shown | Anything wrong about us |
|---|---|---|---|---|---|---|---|
| 2026-10-15 | (assistant name) | Where can I order a wedding cake in Leeds? | No | n/a | Example Cakes, Sample Bakery | example.com/guide | Opening hours were out of date |
One row per try. Ten rows tell you little. Forty rows start to show a pattern.
Side by side
| Asking yourself | AI Pick Check | |
|---|---|---|
| Cost | Free, plus your time | The free check costs nothing. The full audit is $39 once |
| Time | Around twenty minutes for a careful round | About 2 minutes for the free check, several minutes for the full audit |
| Number of answers | As many as you have patience for | Free check: 10 answers. Full audit: 160 answers |
| Same question asked more than once | Easy to forget | Built in, with a margin of error shown |
| Questions without your own name in them | Hard to avoid, because you know your business | Written to sound like a buyer, not like you |
| Who is named instead of you | You note it down | Counted and ranked |
| Which sources the assistants cite | You open each link | Collected and sorted |
| Wrong facts about your business | You may spot some | Checked against your own site |
| What to do next | Up to you | A ranked fix plan tied to what was seen |
| Repeatable later | Hard to compare with last time | Same questions, with one free re-run within 60 days |
What the full audit does differently
The full audit asks 20 questions. Five are buyer needs, each worded three ways. Five more are about your own facts. Every question goes to four AI assistants through APIs with web search, twice each. That makes 160 answers. One of the four, Google's Gemini model, is reached through Perplexity's API and searches with Perplexity's web search, not Google Search. Our methodology page explains this.
One honest note. An API is not the same as the chat window a person types into. The answers can differ from what a chat window shows, so we show ranges instead of single figures. Our blog post on API answers and app answers explains the gap. If you want to see exactly what shows on screen for you, asking by hand is still the better way.
When asking yourself is enough
- You want a quick feel for how assistants describe your business.
- You have a small local business and a handful of obvious buyer questions.
- You are not ready to change anything yet.
When a structured check helps
- You want to know how often you are named, not just whether you were named once.
- You want to see which competitors take the places you expected.
- You want to know which pages the assistants trust, so you can work on those.
- You will make changes and want to know whether they worked.
How to check it worked
If you ran the by-hand routine, check three things before you trust it. Did you use a fresh chat each time? Did you ask each question at least twice? Do your notes show a range, such as "named in 2 of 4 tries", rather than a yes or no? If so, your numbers are fair for what they are: a small sample.
What this won't do
- Asking by hand will not give you a margin of error. Twenty tries are still a small sample.
- Neither route can tell you why an assistant named someone else. It can show you the sources it cited, which is a good place to start.
- Neither route can promise a result. Assistants change their answers, and no check controls that.
- Tools that read the public chat pages are a third route. They are covered on our Otterly comparison.
A fair way to start
Do both. Ask three or four questions yourself first. Then run the free check on the same business and compare. If the two agree, you have learned something. If they do not, that is a good reason to ask more than once.