Ask an AI assistant whether a firm can be trusted and it answers from what the web already says: reviews, profiles, articles and the firm’s own pages. This post draws on reputation work we ran for a UAE law firm, and on the questions we track about our own name, then gives a worksheet for checking yours.
What a law firm’s reputation work taught us about reviews
The firm advises businesses on regulation, licensing, investigations and enforcement, with a focus on the UAE and on virtual assets. Its clients choose an adviser on trust, and much of that trust forms online before anyone makes contact. A prospective client’s first meeting often happens on page one of a search.
We ran the work as a process, not a campaign, in four steps:
| Step | What it covers | Why it matters for AI answers |
|---|---|---|
| Monitor | The firm’s name watched across search results, review sites and social channels | An assistant can only repeat what is out there. Seeing it first means a problem is found by you, not by a client. |
| Assess | Each item judged on its accuracy, reach and risk | Some items are best left alone, and some need a correction where they were published. |
| Respond | Inaccurate or damaging content addressed through the right channel, in a tone that suits a law firm | A measured public reply becomes part of the record any summary draws on. |
| Promote | Accurate, positive content about the firm’s real expertise given the visibility to be found | An answer built from many accurate sources is harder to skew than one built from three. |
One limit on what we can tell you: our record for the firm covers search results, review sites and social channels. We did not measure the firm’s AI answers, so we make no claim about them.
The lesson that carries over is the last step. The case’s own conclusion is that visible expertise is the best defence, and that is the material an AI answer has to work with. Our reputation recovery page describes the same scope.
The question set we wrote for our own name
On 5 October 2026 we wrote 39 questions a buyer might put to an AI assistant about the services we sell, each mapped to the page a good answer should cite. Six are about PixelCrayons by name. One is the review question every firm should test for itself: “Is PixelCrayons legit? What do its reviews say?”
We read the set monthly, as our search standard sets out, and date every reading. Each answer is logged with its date: did PixelCrayons appear, was the mapped page the source, and did the facts match our fact sheet.
There are no results in this post, because there were none to report: on 5 October 2026 our Brand Radar reports held no readings for these questions yet. What we can show now is the method.
A worksheet for checking what AI says about your reviews
You can run a small version of this by hand in an afternoon. Write the questions in your buyers’ words, not yours, and record every answer with its date.
| Question to ask | Page a good answer should cite | What to record |
|---|---|---|
| Is legit? What do its reviews say? | Your trust or reviews page, and your main review profiles | Is the rating current? Are the complaints it repeats real and recent? |
| What is ? | Your about page | Are the description, founding year and ownership right? |
| or : which should I hire? | The page that answers the comparison honestly | Note the links behind each side of the comparison |
| Who offers for ? | The service page | Are you named at all? Who is named instead? |
Ask each question on at least three assistants and repeat monthly. Answers vary between runs, so one reading is an observation, not a verdict. When an answer is wrong, fix the source it drew on, the profile, the page or the outdated listing, rather than arguing with the assistant.
- 01Your own pagesAbout, trust and service pages a good answer cites
- 02Review profilesCurrent ratings, with figures that agree
- 03ReviewsTrue, critical reviews are answered, not removed
- 04Articles and listingsCorrected where they were published, not argued with
How we state our own ratings
The same discipline applies to us. We show our Clutch rating as 4.9 out of 5 from 18 reviews, checked on 15 September 2026, with a link to the Clutch profile. It carries its date because a rating is a snapshot, and we recheck it monthly.
On GoodFirms we link to the reviews without a number. When we checked in September 2026, the profile showed two summary figures that disagree, and an assistant that quotes either one would be repeating an inconsistency. We will add a number when the profile is consistent. If your own profiles conflict, fix them before you worry about AI answers.
What reviews cannot do, and what to avoid
Star ratings are a small part of what an assistant reads, and its weighting is not yours to set. Nobody can promise that an AI answer will recommend you, and we do not.
- Do not pay or reward people for reviews. Google’s review policy does not allow incentives in exchange for reviews, and a burst of rewarded praise undermines the trust it was meant to build.
- Do not expect to remove a true, critical review. Reviews that break platform policy can be reported. The rest are answered. Our guide to removing bad reviews from Google covers which is which.
- Do not treat one AI answer as a ranking. Measure across questions and dates, and read the result as a pattern.
If a reputation problem is already live, monitoring is too slow on its own. Our reputation crisis response covers the first days, when the response matters more than the measurement.
