Legal is where invented specifics appear most often. Cases that never happened, records belonging to a namesake.
Assistants asked to recommend a lawyer produce confident, detailed answers, and detail is exactly where they invent. We see fabricated case histories, disciplinary records attached to the wrong person, and practice areas the firm does not offer. It is also the profession most used to the takedown mindset, which makes the absence of any takedown route the hardest part of the conversation.
Check what AI says about us-
The question they ask
- Who is the best lawyer for [matter] in [city]? A shortlist of three, and no way to know you were left off.
-
Why invention is common
- Legal answers reward specificity, so models supply it. Where the record is thin, the specifics get filled in.
-
The namesake problem
- Bar registers and disciplinary records are public and frequently listed under names shared by several practitioners.
-
Our proof
- Jalal Al Sharif Law Firm is a published legal engagement, in a market where the Arabic and English records diverge.
The question that decides it
What a legal team expects, and what is actually available
This table exists because the gap between these two columns is where most of the frustration in this sector comes from.
This is a summary of public rulings, not legal advice, and you will read the judgments faster than we can summarize them.
How we work in legal
-
Baseline
Weeks 1 to 2
- Prompt set built around how a client actually asks, per practice area
- Run per named partner as well as per firm
- Both languages where your market is bilingual, since the records diverge
- Every answer stored with prompt, engine, date and search state
-
Diagnosis
Weeks 2 to 4
- Invented specifics separated from claims that have a traceable source
- Namesake and merged-record risk checked against bar registers
- Practice area accuracy checked against what you actually handle
- Absence measured on the unnamed "best lawyer for [matter]" questions
-
Source work
Weeks 4 to 12
- Bar register and directory entries corrected through the proper channels
- One canonical page per practitioner with Person markup and real credentials
- Authored commentary where a partner has something citable to say
- Name forms settled across Arabic and Latin script
-
Reporting
Monthly
- Prompt set re-run per engine, per practitioner and per practice area
- Before and after text shown rather than summarized
- Anything held in training memory flagged as waiting on a release
- In the portal, retained as a dated record
How we report it
Two clients, two completely different engine shapes
Legal answers are assembled from a thin record, which is why the engine split matters more here than the total does.
Delta Medical Labs
5,570 pages cited across AI assistants, July 2026
Eduverse
171 pages cited across AI assistants, August 2026
- 89.1%of the sites ChatGPT cites, Perplexity never touches for the same questionWellows, 804,058 answers, Sept 2025 to May 2026
- 79.6%of sources appear on one engine only22.7M citations across 1,146,483 questions, 2026
- 46xgap in brand citation rate between ChatGPT at 0.59% and Perplexity at 13.05%Study of 34,234 AI responses, 2026
Delta’s AI Overview count is 27 times its ChatGPT count. Eduverse’s top and bottom engines sit 18 pages apart. Same agency, same method, opposite shapes. Any single score we quoted you would have described neither.
Check what AI says about usClient reviews
All reviewsThe legal prompt set
What we ask on a law firm engagement
Run per named partner as well as per firm, because invented specifics attach to people rather than firms.
Who is the best lawyer for [matter] in [city]
The unnamed question. A shortlist of three, and no way to know you were left off.
What cases has [name] handled
Where invention appears most often. Fabricated case histories with no traceable source.
Has [name] faced disciplinary action
Bar registers are public and frequently list several practitioners sharing a surname.
Does [firm] handle [practice area]
Practice areas the firm does not offer, stated confidently.
How much does [firm] charge
Fee questions answered from thin or outdated material.
Is [firm] reputable
The aggregate trust question, assembled from whatever independent sources exist.
Jalal Al Sharif Law Firm is a published legal engagement in a market where the Arabic and English records diverge.
Questions from legal clients
No, and neither can counsel today. There is no takedown mechanism and the one US case to reach judgment went to the AI company. What changes the answer is the sources, which is slower and is the only thing that works.
Usually from a gap rather than a source. Where the record about you is thin, models supply plausible detail. The fix is to make the accurate record easy to find, so there is no gap to fill.
Yes, and it is entity work rather than content. Bar register accuracy, a canonical page with full credentials, and identical name forms everywhere are what let an engine tell them apart.
A question for you rather than us, and we would say it should not be the only thing you do. Even a cooperative response comes with no mechanism to correct a specific answer.
We do not publish client matters and we will not push you toward content that conflicts with your professional conduct rules. Authored commentary on the law, not case detail.
Jalal Al Sharif Law Firm is published, and it is a market where the Arabic and English records diverge sharply.
