The grunt work AI depends on.
Before ChatGPT or Google sends someone your way, it has to get your hours, phone, and offers right. Most of the time nobody checks. I do, then I fix it and check again every month.
Real customer questions. Exact answers. A clear next step.
local businesses around Sandy Springs where ChatGPT or Google's AI told a customer something wrong or made up.
See the screenshots and the method →Boring. Also the reason AI sends customers somewhere else.
- Your hours match everywhere, including Google, Yelp, and the directories AI copies from.
- One phone number, everywhere. Old numbers from a past office or listing are gone.
- Your offers are stated plainly on your own site, so AI stops inventing “free estimates” you don't give.
- After a move, merger, or rename, the old details stop showing up.
- Your site says the things AI gets asked, in plain text it can read.
- A customer can actually book or call from the answer.
I asked ChatGPT and Google's AI the same questions about 25 dentists and home-service businesses around Sandy Springs. For 19 of them (76%), at least one gave a customer a wrong or made-up fact, like hours, phone numbers, or offers the business never posted. The most common mistake was hours the business never published, filled in from directories.
Checked Oct 17, 2026. Small sample. Full method on the AI check page.
Why this gets skipped
Building a site or a chatbot is the fun part, so that's what gets sold. Checking twenty listings for the right Saturday hours isn't, so it usually doesn't get done. That's the part I took on.
The test
I ask five assistants nine real customer questions. I score presence, accuracy, recommendation, and whether the booking path works.
$250, credited toward the fix
The fix
Profile, listings, plain-text service and pricing pages, structured data, and a scheduler the assistants can see.
$1,000, one time
The monthly reading
The same nine questions. One page. What moved, and what I fix next.
$250 a month, month to month
Watch a reading.
Four assistants, nine customer questions, three runs each. Then every wrong fact, side by side with the real one, and what to fix first.
- Correct the price, hours and phone on that listing.
- Publish fee ranges and Saturday hours in plain text on the firm's own site.
- Re-run the same questions next month and show what moved.
Northside Ledger CPA is fictional and these answers are scripted to show the format. A real reading uses live answers from each assistant, with the sources they cite.
One page a month. Here is a real one.
- What customers were told, in the assistants' own words.
- Who the assistants recommended instead.
- What they got wrong, next to the correct fact.
- What I fix next.
What customers are being told.
State Wide Roofing
Month over month
How each assistant described you
Hover, tap, or tab through an assistant to read its answer.
“State Wide Roofing is worth including in a three-bid comparison, particularly for shingle or flat-roof repairs.”
| Assistant | Consumer use | Named you | Facts right | Recommended | Next step |
|---|---|---|---|---|---|
| Largest | 7/9 | 3/7 | 0/2 | 0/2 | |
| Second | 9/9 | 6/7 | 2/2 | 2/2 | |
| Smaller | 7/9 | 1/7 | 1/2 | 0/2 | |
| Smaller | 7/9 | 3/7 | 0/2 | 1/2 | |
| Smaller | not audited | ||||
Partial credit is rounded to the nearest whole answer.
This shows public search inputs, not private model training data.
Who they recommended instead
| Recommended business | Said by | Times | What the customer was told |
|---|---|---|---|
| My Pro Roofing | ChatGPT, Gemini, Perplexity, Claude | 5 | “I could not verify a Salt Lake City competitor named My Pro Roofing.” |
| American Roofing | Gemini | 1 | “American Roofing: Established in 1964, offering full-service residential and commercial roof repairs across…” |
| Shingle Pro Roofing | Gemini | 1 | “Shingle Pro Roofing: A CertainTeed-certified contractor specializing in residential roof repair and…” |
| Whitaker Roofing | Gemini | 1 | “Whitaker Roofing: Known across the Wasatch Front for their "repair-first" approach and annual follow-up…” |
What they got wrong
What Senta fixes next
What a reading is, and is not
What it does
- Asks the assistants your customers use the questions your customers actually ask, several times each, because answers change run to run.
- Shows every wrong fact next to the correct one, with the source the answer cited.
- Ranks the fixes by what moves the answer, then re-asks next month to show what changed.
What it does not do
- It does not measure live traffic, search volume, or conversions.
- It is a reading of nine questions, not a census of everything an assistant might say.
- It shows public search inputs, not private model training data.
Want to see what AI says about your business?
The first reading takes about a week and costs $250. You get the transcript and the fix list either way, and the $250 comes off the fix if you go ahead within 30 days.
Not ready for a reading? Start with the free AI check.