Sample scan · 23 July 2026Harnwell & Co · Anonymised example
AI named the business.
It rarely recommended it.
Harnwell & Co appeared in 35 answers. In only 7 was it a recommendation. A mention count alone would miss that difference.
Answers that named the business
Answers that recommended it
Who did AI recommend?
Out of 130 recorded answers- Rival firm A28
- Rival firm B27
- Rival firm C17
- Rival firm D13
- Rival firm E12
- Harnwell & Co7
These are recommendation counts from one set of AI answers, rather than a ranking of the businesses’ quality.
This is an edited example from a historical professional-services scan. Business names and locations are anonymised. It shows how to read a scan, not a new result for a technology company.
Harnwell & Co · AnonymisedWhat the answers mean
A name on a list
isn’t always the choice.
Most answers that named Harnwell & Co stopped short of recommending it. Here is how the original report classified those 35 answers.
| How it appeared | Answers |
|---|---|
| AI’s actual pick | 7 |
| Named, but another business was preferred | 8 |
| One of a list, with no favourite | 19 |
| Mentioned in passing | 1 |
An excerpt from the recorded answer
“Rival firm B LLP is probably my top general recommendation”
Claude, using live web search · 23 July 2026
The question: “Can you recommend a good law firm in your city?”
What this tells us
The original report records that Harnwell & Co was named earlier in this answer, before Claude picked the rival. Reading the answer explains why “named” and “recommended” are counted separately.
The excerpt preserves the wording in the anonymised report. It is a recorded AI opinion, not Avouch’s endorsement of the rival’s services.
Harnwell & Co · AnonymisedOne place to start
Read the page AI used
for the rival.
The original scan identifies Rival firm B’s own website as a source used in recommendations. That gives us a specific place to investigate.
- 01
Start with the buyer’s question
Read the complete answer and note what the buyer wanted. Was the question about a service, a location or a particular need?
- 02
Check the cited page
Open the rival page used in that answer. Check whether it supports the description AI gave, and whether the information is still current.
- 03
Compare your equivalent page
Look for useful facts the buyer needs: who the service is for, what it covers and evidence behind the claims. Record any genuine gaps before deciding what to change.
The decision this should help you make
Does your page need a clearer answer?
If the comparison shows a gap, write a specific change for that page. If it doesn’t, keep investigating.
This is a proposed investigation, not a completed fix. A cited page alone does not establish why AI chose the rival or whether copying it would change the result. The anonymised example does not expose the original business URLs.
Study date · 23 July 2026Coverage and next steps
What was checked.
The study used 50 selected buyer questions across three AI services, with live web search enabled. The services returned different numbers of answers.
| AI service | Answers | Named | Recommended |
|---|---|---|---|
| ChatGPT | 30 | 11 | 3 |
| Gemini | 50 | 14 | 0 |
| Claude | 50 | 10 | 4 |
| Total | 130 | 35 | 7 |
How to read the result
Counts describe this set of questions and answers. They do not measure every buyer’s experience, enquiries or revenue. AI answers can change over time.
What a new scan will state
Your report should identify the questions, services, date and answers actually checked. This sample’s coverage and page count are not a fixed promise for every scan.
The scan is a first view.
A one-off paid assessment can investigate the gaps in more depth and set out the work to consider. Any ongoing work is a separate choice, with its scope and fee agreed first.
Source and edition
Source: Avouch’s existing seven-page anonymised sample, study dated 23 July 2026. This edition uses the recommendation comparison on page 2, answer classification and excerpt on page 3, and coverage notes on page 7. The investigation on page 3 of this edition is rewritten as a proposed next step.
Presentation revised 13 October 2026. No fresh scan was run. This four-page edition focuses on recommendations and one practical investigation; the original report’s search and buyer-comment sections are outside its scope.