Why ChatGPT recommends your competitor's law firm (a footprint audit)

Someone showed you the screenshot, or you typed the question yourself: "best [practice area] lawyer in [your city]" — and there they are. The firm you beat in court last spring, recommended by name, while you're not in the answer at all.

It stings, and the sting is rational. That answer is a private referral, delivered at the moment someone decides to hire, and nobody calls to tell you it happened. But before you do anything about it, do one thing first.

First, verify it's actually real

One screenshot is not a trend. AI answers are unstable by nature: in March 2023, AttorneySync ran the same lawyer-recommendation prompts repeatedly and got "different lists of law firms" each time. Wording alone can flip the outcome. In Carnegie Mellon-led experiments published at CHI 2025 (Lin et al., LLM Whisperer), everyday rewordings of a question changed whether a brand appeared in answers at all — from never mentioned to always mentioned — and when the synonym choices were adversarially optimized, the swing reached up to 78.3 percentage points. Same intent, different wording, different winner.

So run a real sample before concluding anything:

• Ask the question three different ways: the way you'd search it, the way a stressed client would type it ("got a DUI last night in Austin, should I get an attorney?"), and a comparison form ("who are the top 3 family law firms in Columbus?").
• Run each phrasing on at least two assistants (ChatGPT plus Gemini or Perplexity), in fresh chats.
• Log every answer: who was named, in what order, which sources were cited.

Competitor in four or more of six answers? That's a pattern, not a fluke, and a pattern has causes you can find. (If you're absent everywhere regardless of who appears, start with the sibling diagnostic: Why is my law firm not showing up in ChatGPT?.)

Hold onto this: the assistant didn't choose them out of admiration. It assembled an answer from sources, and their sources are better than yours. That footprint is public. You can audit all of it this week.

Now walk through each row.

Their directory profiles are doing the talking

Start where the citations start. A legal directory was the first source cited in 77.8% of AI answers to lawyer-hiring questions (1,254 of 1,612 valid answers, 540 queries asked three times) — InterCore Research, July 2026. InterCore sells AI-visibility services; this is vendor-published research, cited with that caveat. The 5WPR/Haute Lawyer 2026 report separately found seven directories (Chambers, Legal 500, Super Lawyers, Best Lawyers, Martindale, Avvo, Justia) dominating AI citations for legal queries.

Pull up your competitor's Justia, Avvo, and Super Lawyers profiles next to yours. Odds are theirs are claimed and complete — photo, bio, practice areas, office locations, linked reviews — while yours is a stub the directory auto-generated years ago. To an AI assistant citing directories first, a complete profile is the firm; a stub is a rumor.

Someone else keeps writing about them

AI assistants overwhelmingly cite what third parties say. In Muck Rack's May 2026 analysis of AI-cited links, 84% of citations went to earned, third-party media rather than brand-owned pages, and every edition of that recurring study has put the share in the low-to-mid 80s. (The stats hub collects this and every other number on this page, with sources.)

Search "Their Firm Name" -site:theirfirm.example and read the first three pages. The likely pattern: a partner quoted in local news when a big case hit, a bar association article, community-sponsorship writeups, and, most importantly, presence in local "best lawyers in [city]" roundups, among the page types AI assistants cite most. None of that is luck. It's years of saying yes to reporters and showing up in lists.

Their reviews are heavier — and fresher

Review platforms feed both directories and assistants. Compare volume and dates, not stars: 180 Google reviews with ten from the last 60 days reads as an actively chosen firm; 34 reviews that stopped in 2024 reads as dormant, whatever the average says. Check whether they respond to reviews, too. Responses are one more block of crawlable text tying the firm to its practice areas and city.

Their website answers questions; yours describes credentials

Open their practice-area pages. Winning firms tend to have pages that read like answers: "How much does a car accident lawyer cost in Minneapolis?", "Do I need an attorney for a first DUI?" — plain client language, city named, question answered directly, often with FAQ markup that makes the structure machine-readable. If your site leads with "decades of combined experience" and firm history, you wrote for a partner retreat, not for the question being asked.

They're the same firm everywhere

Finally, boring but load-bearing: check their name, address, and phone across every profile above. Consistent everywhere, most likely. If your footprint is split between "Whitfield & Ross LLP" and "Whitfield and Ross Trial Lawyers" with two old addresses in circulation, a model can't confidently assemble you into one recommendable entity. Confidence is what gets a firm onto a three-name shortlist.

What to do in the next 30 days

Realistic sequencing: hygiene first, sources second, measurement throughout.

1. Days 1–3: Baseline. Run your question set (3 phrasings × 2+ assistants), log every answer and citation. You can't close a gap you haven't measured.
2. Days 3–7: Directory hygiene. Claim and fully complete the free profiles (Justia, Avvo); fix or complete Super Lawyers and Lawyers.com if you're listed. Every field, current address, real bio.
3. Days 3–7: Consistency sweep. One canonical firm name and address; correct every listing, your Google Business Profile, and your own contact page to match exactly.
4. Week 2: Restart the review engine. Ask your last ten satisfied clients for a Google review. Respond to every review, old and new. Recency is the signal: make it a standing intake-close step, not a one-time blast.
5. Weeks 2–4: Publish two answer pages. Pick the two questions from your baseline that you lose worst. Write plain-language pages that actually answer them for your city, with genuine FAQs.
6. Weeks 3–4: Start one earned-media motion. Pitch one local reporter or bar publication where you have something real to say. One credible third-party mention beats any amount of self-description.
7. Weekly, ongoing: Re-measure. Same questions, same assistants, logged the same way — the tracking guide covers the full workflow. Watch your share of answers, your slice of the shortlist, not any single response. (This is the number we call Case Recommendation Share.)

Thirty days of this won't dethrone anyone. It removes the reasons you're easy to skip, and starts the trend line that tells you whether the gap is closing.

The honest limits

Read this part before you spend money anywhere.

Nobody can guarantee your firm an AI recommendation. Anyone who promises one is selling snake oil. The models are probabilistic, retrained on their own schedules, and — as the Carnegie Mellon-led work shows — sensitive to wording in ways nobody fully controls, including the vendors who claim otherwise. There is no submission form and no secret schema tag that reliably puts a firm into ChatGPT's answers. And while ChatGPT now sells clearly labeled ad placements (details in the FAQ below), buying one does not put a firm into the organic, unsponsored shortlist.

What's controllable is the source layer: directory completeness, third-party coverage, reviews, answerability, consistency. Improving those raises the odds across thousands of future answers; it scripts none of them. Attribution stays murky too. If you appear more often next quarter, your changes are a plausible cause, not a proven one. Weekly measurement is what separates "we think it helped" from "our recommendation share went from rare to regular."

Anyone selling certainty in this channel is describing a product that cannot exist. What can exist: receipts.

FAQ

Can my competitor be paying ChatGPT to recommend them?

Not for the recommendation itself — but the picture changed in 2026, so it's worth being precise about what is and isn't for sale. ChatGPT now carries paid ads: OpenAI launched ads to US free-tier users on February 9, 2026, opened self-serve ad buying on May 5, 2026, and shipped its Ads Manager on July 22, 2026. Perplexity has run sponsored follow-up questions since 2024. Those are labeled sponsored placements, visually separated from the answer itself. The organic recommendation — the unsponsored shortlist your competitor keeps landing in — is not for sale on any of these platforms, and no ad buy places a firm in it. What competitors can buy is help with the source layer (directory profiles, review campaigns, PR), which influences organic answers indirectly. If their name appears in the answer body without a sponsored label, they earned the placement through sources, not spend.

Did they hire a "GEO agency" to do this?

Maybe, but it matters less than it seems. Everything in the audit above is visible and reproducible whether an agency built it or a diligent office manager did. Audit the footprint, not the vendor. If you do hire help, apply the honest-limits test: anyone guaranteeing a recommendation fails it.

How long will it take to catch up?

Directory and consistency fixes can show up in browsing-based answers within weeks; earned coverage and review mass compound over months. A firm with five years of accumulated press and reviews has a real head start. What you can change fast is going from unmeasured to measured, which most firms, including many current winners, still haven't done.

Should I just copy everything the competitor does?

Copy the structure, not the content. Claim the same directories, match their consistency, learn from which questions their pages answer. Duplicating their content gets you a thin-copy penalty and a weaker version of their entity. The goal is being the best-documented version of your firm, for your practice areas and cities.