The prompt sheet
10 to 20 questions your clients ask, each in several wordings, run on every engine on different days. You receive: the sheet, with question, engine, date and named or not.
AI SEO agency
We ask ChatGPT, Perplexity, Gemini and Google’s AI Overviews the questions your clients ask, write down the date and the exact wording, and ask again because answers change.
Get your prompt sheetHow AI answers work
AI answers are written from sources. ChatGPT, Perplexity, Gemini and Google’s AI Overviews draw on your pages, your listings and the sites that mention you. Google says its AI features need no special optimization beyond a page that is indexed and eligible for a snippet.
What you do, who you help and where. Pages that answer the question in their first lines give an engine a sentence it can quote without the rest of the page.
Hours, address, categories and reviews on your Business Profile and the directories. When every source says the same thing, there is less for an answer to get wrong.
Who recommends you, and for what. When a trade body or a review site names you, an answer has a second source besides your own site.
How it is measured
The same question names different businesses on different days, so one screenshot can’t tell you where you stand. Every number on the sheet has a question, an engine, a run and a date.
| A screenshotOne answer, once | The run logRepeated and dated | |
|---|---|---|
| The question | One question, in one wording | Each question, in several wordings your clients use |
| The engines | One engine, at one moment | Every engine we track, on different days |
| The result | Named or not, once | Named in how many runs, out of how many |
| The sources | No record of what the answer leaned on | Every cited source, run by run |
| Before and after | Nothing to compare it with | The same questions, run again after the work |
The count, the dates and the wording behind any AI visibility number. If those aren’t on the page, the number can’t be checked, and a number that can’t be checked isn’t a result.
What the answers cite
Across a sample practice’s 12 runs, its own site was cited twice. Directories and competitors were cited far more often.
Whether you’re named is half the picture. The sites an answer leaned on tell you where to work next.
In the sample, directories were cited in 9 of 12 runs and the practice’s own site in 2, so listings came first in the plan.
The sample groups sources by type. Your sheet names each site, so you can open it and read what it says about you.
What we know
Five facts the method is built on, each with its source.
In a study of 30,504 AI answers, whether a brand was named changed between one check and the next 9.7% of the time. The study suggests seven runs per question and engine, spread over two to four weeks. So each question runs on each engine on several dates over two to four weeks.
Search Console reports traffic from AI features inside the Performance report, under the Web search type, with no separate line. Search Console alone can’t tell you what Google’s AI said about you, which is why the sheet exists.
OpenAI says sites that block OAI-SearchBot are not shown in ChatGPT search answers. Blocking GPTBot only opts the site out of model training, so the two lines need different decisions.
Google says Google-Extended controls whether your content is used to train Gemini models and to ground Gemini’s answers, and that it doesn’t affect inclusion or ranking in Google Search. Blocking it is a decision about Gemini.
Google says there is no special schema.org structured data you need to add for AI Overviews or AI Mode. We still mark up what is true on the page, because it states your facts plainly to every engine, but we don’t sell schema as an AI lever.
The run log
A sample run log for one buyer question, asked on three dates. Each cell records whether the practice was named, and the last column records what the answer cited.
The question is recorded word for word, with the date and the time in ET, so the same run can be repeated after the work.
Each engine has a cell per run. A business named in one run and missing in the next shows up as exactly that.
The cited source sits beside the result, so the log also says where the answer got its facts, which is where the fixes start.
The scope
Six parts, from the prompt sheet to crawler access, each with what you receive.
10 to 20 questions your clients ask, each in several wordings, run on every engine on different days. You receive: the sheet, with question, engine, date and named or not.
The pages those questions should land on, rewritten so the answer sits in the first lines, in words an engine can quote. You receive: each page, with the question it answers.
Schema markup for your business, services and people, matching what the page says and checked in a validator before it ships. You receive: validated markup, page by page.
Name, address, hours and services made the same on your site, your Business Profile and the directories answers lean on. You receive: every listing checked, with what changed.
Mentions on the sites your sheet shows answers already cite in your field: directories, trade bodies, local press and review sites. You receive: each mention earned, with its source.
robots.txt and server settings checked so the engines can read your pages, and Google’s AI Overviews tracked on your questions. You receive: an access check and the Overview log.
Our stack
The engines come first, then the tools that measure them and the ones that check what they read.
ChatGPT can search the web while it answers, and names businesses with links to the sources it used.
We run your clients’ questions in ChatGPT in several wordings and on different days, and record whether you’re named and which sources it cites.
The method
Every number the work produces can be traced to a question, an engine, a run and a date.
Question list
| # | Question | Wordings |
|---|---|---|
| 1 | implant dentist Phoenix All-on-4 | 4 |
| 2 | All-on-4 cost Phoenix | 4 |
| 3 | best implant center near me | 4 |
| 4 | same-day implants Phoenix | 4 |
| 5 | dental implants financing Phoenix | 4 |
Questions
Keep exploring
More booked calls from the map pack, tagged by source.
What to fix first, ranked by effect and effort.
A site Google can crawl, index and load fast.
Earned links from sites your clients read, each one logged.
Category and product pages built for search, reported in revenue.
Senior advice for in-house teams, in writing.
The profile, five pages and listings a small business needs first.
SEO for large sites, run through your tickets and releases.
SEO for agencies to resell under their own name.
Start here
We’ll run it on ChatGPT, Perplexity, Gemini and Google’s AI Overviews, more than once, and send back the dated sheet.