AI Search Optimization
Be the business the assistant names. When a customer in metro Detroit asks ChatGPT, Gemini or Google's AI Overview who to call, the answer is one name and a reason. The work is being that name, measured by asking, every month.
Talk to usIn short
Three things every business an assistant names has in common: a page that answers the question first, an identity that is identical everywhere, and mentions in the sources assistants read. Built for your Michigan business, and measured by putting your customers' questions to the assistants monthly.
- Answer-first pages with the question as the heading
- One identity across site, listing, directories and structured data
- Earned mentions in the sources assistants cite
- Measured monthly by asking five assistants twenty questions
Why the click is disappearing
The answer box takes half the clicks. It gives them to one name.
Pew Research Center watched 900 people’s real Google use for a month. When an AI summary appeared, 8% of visits went on to click a result. Without one, 15% did. One search in five already produced a summary, and the share is rising.
No assistant publishes its rules. But every business they name has three things in common: a page that answers the question in its first sentence, an identity that is identical on the site, the listing and the directories, and mentions in the sources the assistant treats as evidence. We build those three and then we measure the only way that is honest: by asking the assistants your customers’ questions, every month, and recording who they named.
Illustrative. Assistants do not publish their reasons; these three are what every business they name has in common, and they are the work.
- 8%
- of Google visits that showed an AI summary went on to click a traditional result, against 15% of visits without one
- 1 in 5
- Google searches produced an AI summary in March 2025, across 900 people's real browsing
Pew Research Center, Google users are less likely to click on links when an AI summary appears in the results, July 2025.
In the work
Answer first, one identity, earned evidence, measured by asking
Answer-first pages
The question as the heading, the answer in the first sentence, the depth after it. Assistants lift the passage that answers; a page that builds up to its point for four paragraphs is never quoted.
Entity clarity
One name, one address, one phone number and one list of services, identical on the site, the Google listing, the directories and the structured data. An assistant that finds two versions of you names neither.
Structured data that says what you do
Organization, local business, each service, the questions and answers, the service area by county. The machine-readable layer that lets an assistant state what you do and where without guessing from prose.
The sources assistants read
Reviews, the trade association's member page, the chamber, local press, the directories that feed the models. Earned mentions in the places an assistant treats as evidence, because a claim only your own site makes carries less weight than one somebody else makes about you.
A file for the assistants
A plain description of the business written for machines, at the address they look for it, and crawl access open to the assistants you want to be found by. Closed to the ones you do not.
Measured by asking
Every month, twenty questions your customers actually ask, put to ChatGPT, Gemini, Copilot, Perplexity and Google's AI Overviews. Which ones named you, which named a competitor, which named nobody, and what the answer cited. The report is the transcript.
Getting there
Ask, fix the entity, rewrite, earn, ask again
- 01
Ask first
Twenty questions, five assistants, recorded. The baseline is usually that nobody is named, or that a directory is.
- 02
Fix the entity
The name, address, phone, services and hours made identical everywhere they appear, and the structured data written to tie them together.
- 03
Rewrite the pages
Each service page restructured to answer first, with the questions customers ask as headings and the answers complete in themselves.
- 04
Earn the mentions
The association page, the chamber, the reviews, the local press. The sources an assistant reads about you, one at a time.
- 05
Ask again, monthly
The same twenty questions, the same five assistants, the changes recorded. What moved, what did not, and what is next.
Fitted work
What comes up here, specifically
“Who do I call in Farmington Hills?”
Assistants answer a local question from listings and pages that say the city plainly. A site that says “serving southeast Michigan” is not in Farmington Hills to a machine. Yours names the cities it serves, and the counties, in the words and in the structured data.
The Michigan rule questions
What uncapping does to a tax bill, how a no-fault claim works, what the 20-day notice is. Assistants cite the page that explains a Michigan rule correctly, and a business that publishes that page is the one that gets named alongside the answer.
The Arabic question
Assistants answer in the language of the question. A page in Arabic, for a business in Dearborn, is the page cited for the question asked in Arabic. The same for Spanish in southwest Detroit.
The county line
The service area stated as counties, not a radius, so an assistant does not send someone in Livonia to a company that stops at the Oakland County line, and does not skip you for the city you do serve.
Our own Michigan property tax calculator is the shape of a page assistants cite: one Michigan rule, explained correctly, with the arithmetic shown.
Right time when
When this is worth doing
A good fit
- Customers tell you they asked ChatGPT and it named someone else
- Your service pages build up to the point instead of starting with it
- Your name, address or services differ between the site and the listing
- You already rank and want to keep the click as the results change
Not a fit
- There is no search foundation yet, in which case that comes first
- You want a guaranteed mention, which nobody can honestly sell
- You want pages written for machines, which we will not write
Scope
What we do not do
Stuff pages for machines
No hidden text, no question lists nobody asked, no pages generated to match every phrasing. What reads badly to a person is read badly by the model, which was trained on people.
Fake the evidence
No bought reviews, no invented mentions, no directory listings for addresses you do not have. Assistants weigh the same sources people do, and the same sources catch the same fakes.
Promise a mention
No assistant publishes its rules and every one of them changes monthly. We do the three things every named business has in common and report, from the transcript, whether you were named.
Reasonable to ask
Before you commission it
What is AI search optimization?
The work of being the business an AI assistant names when someone asks it a question you could answer. It goes by other names, generative engine optimization (GEO) and answer engine optimization (AEO) among them, and comes down to the same three things: pages that answer first, an identity that is consistent everywhere, and mentions in the sources assistants read.
Does it replace search engine optimization?
No. It sits on top of it. Assistants read the same pages and the same signals Google does, with a stronger preference for a direct answer and for evidence from other sites. A business with no search foundation has nothing for an assistant to find.
Which assistants do you measure?
ChatGPT, Google Gemini and AI Overviews, Microsoft Copilot and Perplexity, because between them they are where most of the questions are asked. The list changes as the market does, and the report says which were asked.
How do you know it is working?
By asking. The same twenty questions are put to the same assistants every month and the answers are recorded. The report shows which named you, which named a competitor, and what each answer cited, so the change is visible and the next step is obvious.
Can you get us into AI Overviews?
Nobody can promise it. What we can do is the work that every business appearing in one has done: an answer-first page, consistent identity, structured data and earned mentions. Then we measure whether it happened.
Ask an assistant about yourself.
Send us the question a customer would ask and what ChatGPT or Gemini answered. We will tell you why it answered that way, from your pages and your listing, and what the three fixes would be.
Talk to us