# AI answers about doctors and lawyers: what to fix

How AI answers about doctors and lawyers draw on registers, regulator decisions and directories, what you can correct at source, and how to audit them.

---


AI answers about doctors and lawyers are built from the same public material a patient or client would find by hand: the regulator's register, published regulator decisions, directory profiles, review sites, hospital or firm rosters and the professional's own pages. When an assistant gets a named professional wrong, the cause is most often one of three things: a namesake, a stale profile, or an old regulatory outcome that is still public.

A register entry or a directory profile can be corrected at the source, and an assistant that reads the corrected page will usually follow. A published disciplinary decision cannot be edited on request, and it stays visible for a period set by the regulator's own policy. Knowing which is which decides where the work goes.

This guide covers doctors, lawyers, accountants and similar licensed professionals. Rules differ by country and state, so each regulator is labeled with its jurisdiction. It is information, not legal advice: anything that touches a regulator, a complaint or a possible claim belongs with your own counsel or indemnity provider.

<div class="box"><b>The short version</b>
<ul><li>Assistants assemble a professional's profile from registers, regulator decisions, directories, rosters, reviews and the professional's own pages, then fill gaps with guesses.</li><li>The common errors are namesakes, outdated affiliations and addresses, and old disciplinary outcomes presented without dates or context.</li><li>Your own pages, rosters, directory profiles and register details can be corrected at source. A published regulator decision cannot be edited on request, and each regulator sets how long it stays up.</li><li>Audit with the same prompts every time, check every claim against the register, and send nothing to a regulator or a platform until you know which source the error came from.</li></ul></div>

## What AI assistants draw on when asked about a named professional

Assistants read a mix of institutional rosters, specialty or professional body profiles, national directories, registers and review sites, with the professional's own site as one voice among many. In its [Search Central documentation on AI features](https://developers.google.com/search/docs/appearance/ai-features), Google says its AI Overviews and AI Mode may use a technique it calls query fan-out, issuing multiple related searches across subtopics and data sources, so the links they show can be broader than a classic search would return.

One measured example comes from a Medical Economics article by the founder of a vendor that sells AI visibility services to practices, who discloses that relationship. It analyzed answers to roughly 1,200 patient-style prompts across 10 specialties and 20 US metro areas, run in May 2026 against GPT-5.5 through the OpenAI API, producing about 33,000 citation events. The author describes the findings as correlational, says the study covers ChatGPT only and the United States, and notes a selection bias in the sources it surfaced. Treat it as a picture of one assistant, not a rule.

<figure class="keyfacts">
<div class="keyfacts-grid">
<div><b>27.5%</b><span>of citation events pointed to hospital and academic rosters</span></div>
<div><b>22.1%</b><span>pointed to specialty-board or association profiles</span></div>
<div><b>5.1%</b><span>pointed to state medical-board license records</span></div>
</div>
<p class="fig-src">Medical Economics, "Before you pay for AI visibility for your practice, audit these five sources": shares of all citation events; ChatGPT only (GPT-5.5 through the API), US only, correlational, patient-style prompts rather than questions about a named doctor, and from an author who discloses a vendor interest.</p>
</figure>

For lawyers, a bar association article on law firm visibility in generative AI says assistants primarily use law firm websites and lawyer and law firm directory profiles. A lawyer directory publisher, Best Lawyers, adds that most sources feeding AI descriptions of a lawyer sit outside the lawyer's direct control, and names a former firm's bio and a stale directory entry among them.

In that sample the state board record was cited far less often than rosters and profiles, so do not assume an assistant has read the register itself. A regulator decision also reaches an answer indirectly, through a news report or a directory that copied it, and correcting one copy does not correct the others.

## Where answers about regulated professionals go wrong

Answers go wrong in five repeatable ways, and each has a different fix. The first is the namesake. A local radio report from June 2025 describes a health system publicly saying that an AI chatbot had wrongly linked one of its physicians to a violent crime suspect on the basis of a shared surname, with no other connection. The health system said the shared name was purely coincidental and reported the false posts to the platforms involved. Most namesake errors are quieter: a different doctor with the same name in another state.

The second is the stale affiliation. The Medical Economics author reports that a stale address left by a former employer was the most common profile error the researchers encountered, without giving a count. The Best Lawyers piece raises the same risks for lawyers: a former firm's bio, a tool that places a lawyer in the wrong state, or one that credits a specialization the lawyer never claimed.

The third is the old outcome without context: a decision that is still lawfully public, summarized with no date and no sanction type. The fourth is the merged record, where a complaint against a firm, a hearing about a colleague and the professional's own profile become one story. The fifth is the confident gap-fill: no source gives a city or specialty, so the answer supplies one.

Label every wrong statement by its likely cause before acting. A wrong specialty usually comes from a directory, a wrong city from an old roster, and discipline that is not yours from a namesake.

## What you can correct at source and what stays public

You can correct what you or an institution publishes, and you cannot edit what a regulator or court has published. The table sorts the common sources by who controls them. Even where the answer is yes, a correction counts only once the assistant, or the search index behind it, reads the updated page.

<div class="tablewrap"><table><thead><tr><th>Source</th><th>Who controls it</th><th>Correct at source?</th><th>What to expect</th></tr></thead><tbody>
<tr><td><strong>Your own website or firm bio</strong></td><td>You</td><td>Yes, immediately</td><td>Make it consistent with the register: full name, regulator number, workplace, jurisdiction</td></tr>
<tr><td><strong>Hospital, firm or academic roster</strong></td><td>The institution</td><td>Yes, by request to its web team</td><td>The most cited type in the study above</td></tr>
<tr><td><strong>Directory profile</strong></td><td>The publisher</td><td>Usually, through its claim or edit process</td><td>Peer ratings and rankings belong to the publisher, not to you</td></tr>
<tr><td><strong>Review site listing</strong></td><td>The platform and the reviewer</td><td>Only under the platform's policy</td><td>Factual errors in the listing, yes; opinions, no</td></tr>
<tr><td><strong>Register details</strong></td><td>The regulator, with you maintaining them</td><td>Yes, through the regulator's own process</td><td>Keep workplace, contact and status details current through that process</td></tr>
<tr><td><strong>Published regulator decision</strong></td><td>The regulator or tribunal</td><td>No edit on request</td><td>Stays for the published period; early removal only through a formal route</td></tr>
<tr><td><strong>Court judgment or news report</strong></td><td>The court or publisher</td><td>Rarely</td><td>Treat as a permanent record unless a court or publisher acts</td></tr>
</tbody></table></div>

The SRA, which regulates solicitors in England and Wales, shows how this works for register details. Its register page says publishing information in the register is a regulatory requirement, that details must be updated through the solicitor's mySRA account, and that the SRA can hide details on request in rare cases, such as a serious risk of harm. The page also warns that a no-results message does not by itself mean a person or firm is unregulated.

The same rule applies to your own pages. The SRA transparency rules require an authorised body to display its SRA number and digital badge in a prominent place on its website, which is also the clearest anchor tying a site to the right register entry. A US physician can do the same with a license state and number. Google says there are no additional requirements or special optimizations for AI Overviews and AI Mode beyond being indexed and eligible to show in Search with a snippet, and its guidance lists keeping Business Profile information up to date among the checks. Fix the pages first, then confirm they are indexed.

## How long regulators keep decisions public, by jurisdiction

How long a decision stays public depends on the regulator and the outcome, and nothing is uniform. The table states what each regulator's own pages said when read for this guide. Check the policy page before relying on any period.

<div class="tablewrap"><table><thead><tr><th>Regulator and place</th><th>What is public</th><th>How long</th><th>Route to challenge</th></tr></thead><tbody>
<tr><td><strong>SRA, solicitors, England and Wales</strong></td><td>Register entries and published decisions</td><td>Set by seriousness since 2 June 2025; the most serious outcomes stay longer</td><td>Details hidden on request only in rare cases</td></tr>
<tr><td><strong>ICAEW, chartered accountants, England and Wales</strong></td><td>Disciplinary database and hearing outcomes</td><td>For example reprimand 5 years, severe reprimand 7 years</td><td>Name omission only in exceptional circumstances; early removal not ordinarily granted</td></tr>
<tr><td><strong>Ahpra, health practitioners, Australia</strong></td><td>National register, with cancelled and undertaking lists</td><td>Not stated on the page read</td><td>Reprimands: application to the National Board</td></tr>
<tr><td><strong>GMC, doctors, UK</strong></td><td>Register status, training and other information; searchable by GMC number</td><td>Policy page could not be read for this guide</td><td>Not checked; see the GMC directly</td></tr>
<tr><td><strong>US state medical boards</strong></td><td>License and board actions, as each state decides</td><td>Not checked for this guide; set by each state</td><td>The state board</td></tr>
<tr><td><strong>US state lawyer discipline</strong></td><td>Public once formal charges are filed, in the ABA model</td><td>Not checked for this guide; set by each state</td><td>The state disciplinary agency or court</td></tr>
</tbody></table></div>

On the SRA, its [June 2025 publication policy update](https://www.sra.org.uk/news/news/sra-update-140-publication-policy/) says the new approach took effect on 2 June, that most decisions were previously published for a fixed three years, and that publication lengths now depend on the seriousness of the misconduct. The most serious outcomes stay public for longer than under the old approach, and decisions are very rarely left unpublished.

The [ICAEW publication of decisions policy](https://www.icaew.com/-/media/corporate/files/policies/icaew-publication-of-decisions-policy-1-feb-2026.ashx), in the version read for this guide, is the most detailed. It publishes a period for each sanction: no further action one year, a caution four years, a reprimand five years, a severe reprimand seven years, and a suspension for the length of the suspension plus five years. Names are omitted only in exceptional circumstances that outweigh the public interest, which the policy says should be granted sparingly. Early removal is not ordinarily granted: the policy names an overturned decision, an error that makes the entry misleading, or exceptional circumstances.

On Ahpra's [removal of reprimands page](https://ahpra.gov.au/Registration/Monitoring-and-compliance/Removal-of-reprimands.aspx), a reprimand is a formal and public denunciation of conduct. The National Board considers a removal application and looks at whether relevant events occurred during the publication period. The page gives no fixed waiting period.

In the United States, the ABA Model Rules for Lawyer Disciplinary Enforcement are a template that each state adapts. Under [Rule 16](https://americanbar.org/groups/professional_responsibility/resources/lawyer_ethics_regulation/model_rules_for_lawyer_disciplinary_enforcement/rule_16) as written, a proceeding is confidential until formal charges are filed and served, and public after that. For US health care providers, the National Practitioner Data Bank describes itself as a confidential information clearinghouse, and practitioners can check reports about themselves through its Self-Query. We could not read the Federation of State Medical Boards DocInfo page or any state board policy for this guide, so state medical board and state lawyer discipline periods are not stated here: check each board directly.

## How to stop an assistant confusing you with a namesake

An assistant cannot reliably tell two people with the same name apart unless the pages it reads carry a shared identifier and the same facts. The task is to make every page about you agree, and to put an identifier on each of them.

The identifiers exist already. The GMC register can be searched by GMC number, and the Ahpra register by name or registration number. In the United States, CMS says it assigns National Provider Identifiers (NPIs) through its enumeration system, NPPES, and makes the data available in downloadable files and a query-only database. Use your regulator's identifier, in the same format, on your site, your roster page and your claimed directory profiles.

<aside class="callout">
<span class="callout-label">Rule</span>
<h4>One name format, one identifier, one workplace everywhere</h4>
<p>Pick the exact form of your name, your regulator number and your current workplace, and use them on every page you control. Where a page you do not control disagrees, correct that page first.</p>
</aside>

Add facts that separate you from the other person: jurisdiction, specialty and employer. Then test: ask each assistant about the person with your name in your city, then in the other person's city, and see whether the answers stay apart.

If the namesake is the one with the disciplinary record, the aim is not to hide it. It is to make sure your pages cannot be read as the same person. That entry is accurate, public and not yours to challenge.

## How to audit AI answers about doctors and lawyers, step by step

Audit the answers with the same fixed prompts every time, record the sources each assistant gives, and compare every claim with the register. The method works for any licensed professional, and the first pass usually takes an hour or two.

<figure class="fig">
<figcaption><b>Audit of AI answers about a regulated professional</b></figcaption>
<ol class="fig-steps">
<li><b>Open the register first</b>Save a dated copy of your own entry, and of entries for people with your name. Every answer is judged against it.</li>
<li><b>Fix the prompts</b>Write six to eight prompts and keep them unchanged: who is this person, where do they practice, what are their qualifications, are there regulatory findings, what do clients say, and who else has this name.</li>
<li><b>Run them on each assistant</b>Use ChatGPT, Gemini, Perplexity, Claude, Microsoft Copilot and Google AI Overviews, each in a fresh session, and note the date.</li>
<li><b>Record the sources shown</b>For each statement, note the page the assistant cites. If it cites none, mark the statement as unsourced and treat it as a guess.</li>
<li><b>Label each error by cause</b>Namesake, stale affiliation, old outcome without context, merged record or gap-fill. The label decides where the fix goes.</li>
<li><b>Fix at source, then re-test</b>Correct the page that holds the error, wait for a re-crawl, and run the same prompts again.</li>
</ol>
</figure>

Answers vary from one run to the next, so run each prompt more than once: a claim that appears once in five runs is a different problem from one that appears every time. Our [AI search reputation service](/ai-search-reputation) covers what assistants say about you. When a statement about a regulatory finding does not match the register, keep the screenshot, prompt, date and register extract together, because your counsel, indemnity provider or a provider's reporting form will ask for them.

## A worked scenario: one namesake, one stale profile, one old decision

This scenario is invented, with generic parties. A solicitor in England practices under a common name. An assistant says she works at a firm she left two years earlier, practices in a city she has never worked in, and was fined by the regulator.

The audit finds three causes. The old firm's website still lists her. An unclaimed directory profile shows the other city, taken from a different solicitor with the same name. The fine appears in a published decision about that other solicitor, and the assistant merged the two profiles.

The fixes go to three places. She asks the old firm to remove her bio and updates her register details through her mySRA account. She claims the directory profile, corrects the city and adds her SRA number. She adds her number and current firm to her own site and keeps her name identical on every page. She does not contact the regulator about the other solicitor's decision, because it is accurate, public and not hers.

<div class="tablewrap"><table><thead><tr><th>Stage</th><th>What happens</th><th>Usual range (experience)</th></tr></thead><tbody>
<tr><td><strong>Audit</strong></td><td>Register copy, prompts, sources, error labels</td><td>A few hours over one to three days</td></tr>
<tr><td><strong>Source fixes</strong></td><td>Old bio removed, directory claimed, own pages aligned</td><td>One to three weeks, depending on third parties</td></tr>
<tr><td><strong>Re-crawl</strong></td><td>Search indexes pick up the corrected pages</td><td>Days to a few weeks</td></tr>
<tr><td><strong>Re-test</strong></td><td>Same prompts, same assistants, compare with the first run</td><td>Two to six weeks after the fixes</td></tr>
</tbody></table></div>

These ranges are experience, not promises. Assistants that search the live web tend to reflect a corrected page sooner than those answering from training data alone.

## The care needed around anything that touches a regulator

Treat anything that touches a regulator as a separate track from the AI audit, with its own advisers and its own timing. A wrong sentence from an assistant is a reputation problem. A published decision, an open complaint or an investigation is a regulatory matter, and what you say about it can become part of it.

<aside class="callout">
<span class="callout-label">Warning</span>
<h4>Never contact a complainant or pressure a regulator to change an AI answer</h4>
<p>Approaching a patient or client who complained, or asking a regulator to alter an accurate public entry because an assistant repeats it, can create a second and worse problem. Use the regulator's formal routes only, and take advice first.</p>
</aside>

Four rules keep the work safe. Separate the questions: the assistant's wrong statement goes to the provider and the source page, while a published decision stays with the regulator's process. Use only the formal routes the regulator documents, such as an Ahpra reprimand removal application or an ICAEW early-removal request on the grounds its policy lists, and expect refusal unless the grounds are strong.

Correct, never conceal. Pushing an accurate, lawful public outcome out of sight with misleading profiles is the wrong goal. Publishing accurate context, such as the outcome, its date and what has changed since, is lawful and usually more persuasive.

Keep your register details true, and do not describe a sanction as lifted or expired on your own pages unless the regulator's record says so.

Take advice before any statement about a pending matter. Your counsel or indemnity provider decides what you may say publicly, and whether a defamation or data protection route applies. Patient and client confidentiality also limits what you can say in reply to a review.

## Reviews, testimonials and advertising rules

Review sites feed assistants too, but they made up a small share of citations in the study above, so a profession's review rules matter more for compliance than for AI visibility. Ahpra's testimonial tool, for example, says the National Law does not allow testimonials to advertise a regulated health service in Australia, and tells advertisers to take care if they engage with reviews on third-party sites. That page shows a review date of August 2021, so check Ahpra's current advertising guidance before acting.

For the audit, a listing can be wrong in fact, such as the wrong address or specialty, while the reviews themselves are not yours to edit. Correct the listing through the platform's claim process, and check your regulator's advertising rules before replying or republishing anything.

## Start with the register, then the answers

Open the register, run the fixed prompts and label each error before you change a page. The register shows what is true, the answers show what the world sees, and the gap is the list of jobs.

<div class="box hot"><b>If this is happening to you now</b>
<p>If an assistant is attributing a disciplinary finding, a complaint or a specialty to you that is not yours, send us the prompts, the answers and the register extract and we will map which source each error came from. Start with a <a href="/audit">private audit</a>.</p></div>

## Questions people ask

### Can an AI assistant show a disciplinary decision that has been removed from the register?

Yes. A removed entry can survive in a cached page, a news report, a directory that copied it or a training dataset. Check each cited page, ask its owner to update it, and keep the regulator's removal notice. An assistant that searches live usually follows a corrected page sooner than one answering from stored knowledge.

### Why does the assistant give the wrong specialty or the wrong city?

Usually because it is reading a directory profile or an old roster, or has merged you with a namesake. The audit shows which page it cites. Fix that page, add your regulator number and current workplace, and rerun the prompt after a re-crawl.

### What should I do if an assistant says I was disciplined and I was not?

Compare the statement with the register entry and save both. If the register shows nothing, the answer is a false statement or a namesake error. Report it to the assistant provider with the evidence, and ask your counsel whether any further step is appropriate. Do not contact the regulator about someone else's entry.

### Is it safe to use an AI assistant to check a doctor or lawyer before hiring one?

Use it as a pointer, not proof. Open the regulator's register directly, search by registration number, and read any decision in full. An assistant may omit a decision, merge two people with the same name or repeat a stale profile, so the register is the source.

### How often should the audit be repeated?

Repeat it after any change to a register entry, a move to a new workplace or a published decision, and otherwise at a steady interval such as quarterly. Keep the prompts identical so differences reflect the answers, not the questions. Reputation Partners can run the cycle monthly where a case needs it.