When a Chatbot Confers the Award: AI Attorney Rankings
Two influence titles on an attorney's homepage trace back to an anonymous ChatGPT session. What the transcript says, and how to verify any credential.
Attorney Blake Harris· Florida Bar #86486, Colorado Bar #45942
Two of the recognition banners on attorney Steven J. Oshins's homepage are not recognitions. They read "Named 'Most Influential [Asset Protection Attorney] Overall Today'" and "Named 'Most Influential Estate Planning Attorney in Nevada.'" Click either one and it opens a PDF, and each PDF is a transcript of a ChatGPT conversation with an anonymous user. No organization conferred those titles and no peers were surveyed.
We think this is worth documenting carefully, because AI-sourced credentials are going to become common and most people will never click through to see where a title came from.
What the Banners Actually Are
The homepage stacks seven banners. Five point to real recognitions: Best Lawyers "Lawyer of the Year" designations across eight years, the NAEPC Estate Planning Hall of Fame (2011), a Wealth Advisor listing, and features in California Business Journal and Law Dragon. Two point to ChatGPT transcripts. They are the second and third items in the stack, styled identically to the rest.
To be fair on one point: each PDF states on its own first page that it is "a copy of a conversation between ChatGPT & Anonymous," copied word for word, with yellow highlighting added afterward. The source is disclosed to anyone who opens the document. Our concern is with the banner, not the PDF - a visitor scanning the homepage sees "Named 'Most Influential'" in the same typography as the Hall of Fame induction, and only a click reveals that the namer was a chatbot.
One further detail is visible in the banner itself. In the first title, the words identifying the practice area appear inside square brackets - "Most Influential [Asset Protection Attorney] Overall Today" - the conventional signal that words have been inserted into a quotation rather than quoted from it.
What the Transcript Says When You Read the Whole Thing
This is where the document undercuts the banner it produced.
The model's first answer refuses the premise. It opens: "There isn't a single universally agreed 'most influential' asset protection attorney." It then lists seven practitioners across four categories - Barry Engel and Jonathan Blattmachr as foundational figures, Richard W. Nenno and Gideon Rothschild as offshore pioneers, Steven J. Oshins and Dan Rubin under domestic asset protection trusts, and Jay Adkisson on the litigation side.
Asked who is "most influential," it splits the answer. Historically, Barry Engel. Technically and in modern planning evolution, Oshins is "arguably at or near the top." That is a qualified statement inside a hedged answer.
The superlative arrives only after a follow-up. The user then types "Rank them in all three areas," and the model produces three separate rankings before synthesizing them. The banner phrase comes from that synthesis, not from the model's own initiative.
And the rankings are not uniform. He places first in legislative influence and first in planning innovation. But under the heading the model itself labels "Impact on Case Law (what actually holds up in court)," he ranks fourth - behind Jay Adkisson, Barry Engel, and Gideon Rothschild. For someone choosing an attorney because they expect a structure to be tested by a creditor, that is the most relevant of the three categories, and it is the one the banner does not mention.
Why an AI Answer Cannot Function as a Credential
Set the specific document aside; the general problem applies to anyone who tries this.
A large language model does not assess competence. It has not read a docket, reviewed a trust deed, or spoken to a client. It predicts the next likely word from text published on the internet, so the name it returns reflects who appears most often, in the most favorable contexts, in its training data. An attorney who has published prolifically about his own work has contributed heavily to that pool. The output is not independent confirmation of anything.
These systems are also built to be agreeable. Ask for a single winner and you will usually be handed one, whether or not the underlying reality supports the premise - which is exactly what the follow-up prompt in this transcript produced.
Nothing about the result is reproducible. Change a few words, ask tomorrow, or ask a different model, and the answer moves. We ran the same question - "who is the most influential asset protection attorney?" - and got the same opening refusal, followed by four names: Jay Adkisson, Barry Engel, Gideon Rothschild, and Richard W. Nenno. Oshins did not appear in our answer at all.
A designation that depends on who asks, how they phrase it, and which model they use is not a designation. And because the sessions are anonymous and undated, there is no way to know how many times a question was asked before the useful answer arrived. Nobody can check the work, which is the opposite of what a credential exists to do.
Contrast that with how legitimate recognitions operate. Best Lawyers, Chambers, and the NAEPC publish criteria, survey peers, and attach an organization's name to the outcome. You can disagree with a result, but you can examine how it was reached and ask someone to defend it. A chatbot has no methodology and cannot be cross-examined.
Why We Are Writing About This Firm Specifically
We have covered this attorney's work once before. His "45 FAPT Cases Gone Wrong" list claims to catalog forty-five decisions proving foreign asset protection trusts fail, and our attorneys reviewed every entry. Two were duplicates, two could not be located, several involved no trust at all, and none showed a properly formed, timely funded trust with an independent trustee defeated on the merits. After we published, he responded on LinkedIn and acknowledged, in his own words, "I haven't read them all," and that "some of them aren't actual trust cases."
In our opinion, the two episodes share a shape: a claim presented with more authority than its underlying source supports, where the gap is visible only to someone who goes and reads the source. A reasonable consumer who reads "Named Most Influential" assumes a person or an organization did the naming.
How to Verify Any Attorney, Including Us
Everything below can be confirmed by you, without relying on a ranking of any kind.
| What to check | How to check it |
|---|---|
| Bar licensure and standing | Look the attorney up in the state bar's own directory, such as the Florida Bar member directory - not through a link the firm provides |
| What the practice actually does | Ask what share of the work is asset protection, and how often they build the structure you need |
| The structure and jurisdiction | Ask which statute governs it, and why that jurisdiction rather than another |
| The trustee | Get the name before you pay, and verify its license with the regulator |
| The total cost | Setup, annual maintenance, and required tax filings, in writing |
None of those requires an adjective. The longer version is in how to choose an asset protection attorney, and the structural reason a licensed attorney matters at all is in privilege versus confidentiality.
The Bottom Line
A recognition is only as good as the person or organization willing to stand behind it. When the source is an anonymous chat session that cannot be reproduced or audited, and whose full text ranks the subject fourth on the measure that decides real cases, the accurate description is not "named most influential."
We asked the same model the same question and it told us no one holds that title. We took it at its word.
Apply the checklist above to any firm you are considering, including Blake Harris Law. We publish our attorneys and their bar numbers, our trustee relationships, and our fees so that you can.
Frequently asked
Frequently asked questions
No. A large language model does not evaluate attorneys. It has not read a docket, reviewed a trust deed, or spoken to a client. It predicts likely text from material published online, so the name it returns reflects who has published and been written about most, not who performs best. Ask the same question twice and you can get two different answers.
It is not a credential in the ordinary sense. Recognitions from organizations like Best Lawyers, Chambers, or the NAEPC Estate Planning Hall of Fame publish criteria, survey peers, and put a name to the result, so you can examine how the outcome was reached. An anonymous, undated chat session has no methodology, no accountability, and cannot be reproduced or audited.
Check bar licensure and standing directly in the state bar's own directory, not through a link the firm sends you. Ask what the firm does day to day, what structure and jurisdiction it recommends, who the trustee will be, and what the total fee is in writing. Each of those can be confirmed by you without relying on any ranking.
Because these systems learn from published text. An attorney who publishes constantly appears more often in the training data, so the model is more likely to surface that name. That makes AI output closer to a measure of publishing volume than of legal skill, which is a poor basis for choosing counsel.
Verifiable facts: an active bar license you confirmed yourself, a practice focused on the structure you actually need, the trustee named before you pay, and complete lifetime pricing in writing. A firm that will not provide those plainly is telling you something more useful than any award banner.
← All articleshttps://blakeharrislaw.com/articles/ai-generated-attorney-rankings