By JayOctober 2026
AI Visibility Takes Weeks. Start With a Baseline.
A customer asks ChatGPT or Gemini who to call, and the answer arrives before your website is ever opened. In early October 2026 we put buyer-style questions to both products for three service businesses in a town north of Toronto. Asked by name, all 6 answers found the right website and the phone number published on it. Asked the way a stranger would, the three businesses were named in 0 of 36 answers. The 36 is 2 questions for each business, on 2 products, repeated 3 times. That is a one-time snapshot from a small pilot, not a before and after.
Asked by name, all 6 answers found the website and the phone number on it. Asked the way a stranger would, 0 of 36.
LinkedIn's 2026 guide (Unlocking AI Search Visibility: The B2B Marketer's Guide to LinkedIn) is written for B2B marketing teams with a department behind them. Its best advice fits a ten-person business too: wait at least 30 days, measure against a fixed set of buyer questions, and judge results over multi-week windows. Our method re-measures at 30 and 60 days for the same reason.
The rest is built for the same kind of team: task forces, creator programs, weekly LinkedIn articles and tracking platforms. A local contractor or a salon can leave all of it out.
Asked by name is not asked by a stranger
A blind question describes what a buyer needs without naming anyone. A branded question names the business. Checking your own name only tests the second kind.
A search for your own name tells you very little about whether a stranger finds you.
One business did the opposite, and its questions were written around its own services. We asked a Toronto salon 3 blind questions about its services, 3 times each, on 4 products. It was named in:
- 9 of 9 blind answers from ChatGPT
- 9 of 9 from Gemini
- 1 of 9 from Claude
- 3 of 9 from Grok
Asked the same question three times in fresh chats over about 15 minutes, Claude named the salon in 1 of the 3 answers, so the same question gave different answers on different runs. Had you checked Claude only once, you could have seen either answer. Grok named the salon on every run of one question and on none of the other two, so for some products the wording of the question changes the result.
A single check can land anywhere in that spread. A baseline is a fixed set of buyer questions that you ask the same way, across several products, more than once. A change then shows up against a range, not one answer.
Make your facts agree everywhere
One of the guide's reasons for the 30-day wait is that AI products need several sources that agree before they treat something as settled. For a local business the first job is to make sure your own sources agree.
Our pilot showed what disagreement looks like. A trades business had one phone number in its site's structured data (the hidden business details that machines read) and a different one on its page and in its association listing. Both products gave the number shown on the page, which matches the association's listing. One of the 6 by-name answers about the three town businesses gave the association's own mailing and office addresses as if they belonged to the business. We also asked about the salon by name, including its opening hours. Of those answers, 4 covered hours, and 1 of those 4 gave hours that differ from the salon's own page.
Every place your facts disagree is a place a model could pick the wrong one.
Where to look, and what to keep the same:
- Your site, with these facts the same on every page and in your structured data:
- Name
- Phone number
- Address
- Opening hours
- Services
- Directory and association listings that carry the same facts as your site.
- Reviews where your customers already look.
- Your own public profiles. If you have a LinkedIn profile, make it match your site.
- One clear description of what you do (in your customers' words), repeated everywhere.
How to Make Your Website Visible to AI walks through the setup, and We Ran Our Own Site Through a Stranger's Checklist shows what we found on ours.
There is no formal standard to lean on yet (The Standards Aren't Written Yet), so the dependable move is consistency.
How long it takes
The guide says to "allow at least 30 days before evaluating AI visibility impact." It gives three reasons for the lag:
- AI products refresh their sources on their own schedules.
- They need several sources that agree.
- They update independently, so visibility moves in steps.
The guide also reports data from Profound (page 28 of the guide) on about 900 new pages cited by ChatGPT or Claude agents between March and May 2026. The guide labels 6.81 days the median, and nine in ten were cited within about 37.
The guide's median is 6.81 days.
1 in 10 cited pages took longer than 37 days.
That covers only pages that were eventually cited, so read it as a best case. Reading the AI Visibility Numbers has the detail.
30 days to start, 60 to confirm
The guide's closing page gives four steps: one core theme, one in-depth article and two supporting posts, two or three credible voices, and a baseline with a review at 30 days. The schedule and the sizes below are ours, scaled to an owner with no staff. We budget about 2 hours per round of questions and about 3 hours a week on steps 2 and 3 (our estimate).
- Week 1: take a baseline. Write 7 questions a buyer would type, with your business name left out. Ask each one twice in ChatGPT, Gemini and one more product, each time in a fresh temporary or logged-out chat so your own history does not shape the answer. Also look at what Google's AI answers and your Maps listing show for the same questions, since local customers use them too.
- Weeks 1 to 2: make your facts match. Align your name, phone, address and hours across your site, your Google Business Profile, Facebook, Yelp or other directories, and any association or chamber-type listing. This alone can take a week.
- Weeks 2 to 4: publish and repeat. Add a page on your own site, such as an FAQ of 300 to 500 words that answers the kinds of questions your customers ask (in their words), plus a couple of short follow-up posts. You and one or two people you work with then say the same thing in the same words on your profiles. Leave the question set alone.
- Day 30 and day 60: re-measure. Ask the identical questions the identical way. In the guide's data one in ten cited pages took longer than 37 days, so day 60 is a second checkpoint.
Record each answer in a simple grid with one row per answer and four columns: question, product, run, and named. Named means your business name appears in the answer, not only a competitor or a directory. Count how many answers name you and compare the totals at day 30 and day 60.
Leading indicators are the things you can see sooner, such as Business Profile views and calls, or whether a listing now shows the corrected number. Outcomes are whether you are named in answers to your fixed questions, and they move slowly, so compare at day 30 and day 60 and not week to week.
Where addAI comes in
This is what addAI.dev does. We set the baseline with a fixed set of buyer questions across products, fix the facts and structure on your site and listings, and re-measure on a schedule. The free scanner is a quick first look at whether AI crawlers (the programs that read websites for AI products) can reach and read yours.
A baseline turns a hunch about AI visibility into a number you can track.