Four AI Platforms Recommended Divorce Lawyers. They Agreed on Almost Nothing.

Someone looking for a divorce lawyer in Nashville this week would get a different answer depending on where they asked. Not a slightly different answer. A fundamentally different one. Different firms. Different evidence. Different reasoning for why each firm deserved to be named at all.

That is the finding that sat at the center of a 175-observation audit published in the first issue of How Machines Decide, a research newsletter that tracks how AI platforms recommend professional services. The audit ran the same real searches Nashville residents actually type through ChatGPT, Perplexity, Google, and Claude on the same day and recorded every firm named, every source cited, and every position held.

50 unique firms surfaced. Three of the four platforms named lawyers. They barely overlapped.

Only five firms appeared on all three. Thirty-four firms appeared on only one. The firm that held the number-one position on Google's Local Pack for every hiring-intent search did not crack the top three on ChatGPT. The firm ChatGPT named first was absent from Perplexity entirely. And Claude, which processes millions of conversations daily, named zero firms for Nashville family law on this date. It was not a referral channel at all.

This is not a story about one platform getting it right and the others getting it wrong. This is a story about three systems building three completely separate trust hierarchies from three different pools of evidence, and the firms caught in between having no idea which version of their reputation is reaching which prospective client.

Modern visibility is not one thing. It is the product of at least three forces operating simultaneously and independently. Search platforms crawl and rank based on technical signals. AI systems read, interpret, and synthesize content to produce editorial-style recommendations. And human behavior filters everything through trust patterns, brand recognition, and emotional resonance. These forces rarely agree, and optimizing for one while ignoring the other two produces the exact kind of fragmentation the Nashville data surfaced.

Consider the 68% of firms that appeared on only one platform. That number is not a failure of marketing execution. It is the predictable outcome of building for one system and ignoring the rest. A firm that invested heavily in Google Business Profile optimization and review generation showed up in Google's Local Pack. But Google Business Profile signals do not feed ChatGPT's recommendation logic. And review volume, the signal that drove Local Pack positioning, did not show up as a factor in Perplexity's citation behavior at all. The investment moved one system and left the other two untouched.

The reverse was equally true. A firm with a prominent Best Lawyers profile showed up when ChatGPT cited legal directories. But that same directory listing did nothing for Perplexity, which pulled nearly half its citations from firm websites rather than third-party directories. And it did nothing for Google's Local Pack, which runs on its own signal set entirely.

Each platform reads from a different shelf. A firm that stocks only one shelf shows up in only one place. The Nashville data showed this happening across all 50 firms at scale.

The citation data is where the audit moved from interesting to structurally important.

ChatGPT surfaced 15 unique firms across 45 findings and behaved like an opinionated colleague who reads legal rankings for a living. It named five firms per answer on average and graded them with review counts, star ratings, and credential citations. It was also the least transparent platform in the study. ChatGPT cited sources only 27% of the time. When it did, 50% of those citations pointed to legal directories like Best Lawyers and Super Lawyers. Another 33% pointed to firm websites. The rest scattered across news and other sources.

The firms ChatGPT favored had peer-recognition credentials. Best Lawyers designations, AAML Fellowships, Board Certification. ChatGPT treated these as authority markers, the equivalent of an endorsement from a system it had already decided to trust. It was not ranking pages. It was synthesizing reputational evidence from sources it deemed authoritative and making editorial judgments about which firms deserved to be named.

Perplexity cast a wider net. 29 unique firms across 45 findings, often listing seven to ten firms per answer. But the structural difference was the citation rate. Perplexity cited sources 100% of the time. Every single recommendation carried an inline citation. Those citations split nearly evenly between firm websites at 47% and legal directories at 47%, with a small share going to news and other sources.

The firms Perplexity favored had practice-area pages that functioned as answers. Fee context. Jurisdiction-specific process breakdowns. Attorney bios tied to outcomes rather than credentials alone. Pages that listed accomplishments without explaining what a prospective client should expect were passed over in favor of whatever third-party directory mentioned the firm instead. Perplexity was reading firm websites the way a researcher reads primary sources, and it rewarded the firms whose sites gave it something worth citing.

Google surfaced 27 unique firms across 85 findings and operated in three distinct layers that rarely agreed with each other. The Local Pack, organic results, and AI Overviews each told a different story. One attorney held position one in the Local Pack on every hiring-intent search. No other firm held that position more than once. Below the Local Pack, organic results shifted firm ordering with every phrasing variation. And AI Overviews, which triggered on 60% of searches (6 of 10), pulled from a noticeably different pool than both the Local Pack and ChatGPT's answers.

Google's citation rate was 0%. It named firms but never pointed to where it found its information. Compare that to Perplexity's 100% and the gap becomes a structural observation about how these platforms relate to their own sources. Google treats its evidence as invisible infrastructure. Perplexity treats its evidence as the product.

This is what it looks like to build content that serves multiple systems simultaneously. Not as a compromise, but as a recognition that each system needs something different. Google needs structured data and local signals. ChatGPT needs directory credentials and reputational consensus. Perplexity needs answer-shaped content on firm websites. A firm that builds all three is building infrastructure that compounds. A firm that builds one is building a wall that only one system can see.

The audit tested multiple phrasings of similar searches. The results broke the idea of a stable ranking more completely than any algorithm update ever has.

On ChatGPT, "best divorce lawyer in Nashville" produced credential-heavy practitioners. Best Lawyers recognition, AAML Fellows, Board-Certified specialists. Swap to "divorce lawyer near me" on the same platform and the results shifted to review-volume firms. Same platform. Same city. Same day. Different sorting logic triggered by a single word change.

Google's AI Overview behavior added another layer. AI Overviews fired on "best" and "top rated" searches but did not fire on "affordable," "custody lawyer," or "near me." The searches that most directly signal someone ready to hire did not consistently produce an AI-mediated answer. When no AI Overview appeared, the Local Pack and organic results were the only answer. When one did fire, it drew from a different source pool than the Local Pack sitting right below it. A single search result page contained two recommendation systems that disagreed with each other.

The high-net-worth search produced the starkest split. ChatGPT cited firms sourced through Chambers rankings. Perplexity cited different firms sourced through Doyle's Guide. Google's AI Overview cited a third set. For the highest-value client segment in family law, the three platforms offered almost zero overlap. Three different answers. Three different evidence bases. Three different firms positioned as the best option for the same prospective client.

A single search intent now travels across multiple platforms simultaneously. The Nashville data made visible what happens when those parallel paths return completely different results. A firm's visibility is not a position on a list. It is a condition that changes with every word typed and every platform consulted. And most firms have no visibility into which version of that condition their prospective clients are experiencing.

The 10% of firms that appeared on all three platforms had something in common. Not one thing. A combination.

Directory credentials feeding ChatGPT. Well-structured website content feeding Perplexity. Strong review profiles feeding Google's Local Pack. And underneath all of it, traditional search optimization holding up the foundation that made AI Overviews possible when they triggered.

That combination creates a flywheel. A firm with strong search optimization earns Google visibility, which builds the indexed authority that AI systems reference, which increases the likelihood of being cited in AI-mediated answers, which drives branded searches that strengthen the search signal further. Success across one system reinforces success across the others. That flywheel effect is what made those five firms visible everywhere while 34 of their competitors were visible in exactly one place.

The difficulty of replicating that combination is the moat. A competitor cannot catch up by investing in one channel. Directory coverage, site-content depth, review strength, and search optimization each move one or two platforms. No single investment moves all three. A firm that starts building across all three systems today will still be months away from the kind of cross-platform presence the data showed, and by then the firms already compounding will have pulled further ahead.

SEO is the foundation of this picture, not the ceiling. The firms with strong traditional search optimization had the base layer that let AI Overviews pick them up when those overviews triggered. The 60% AI Overview trigger rate means search optimization is now feeding two layers of the same platform. Firms that neglect it lose visibility on Google's organic results and on its AI layer simultaneously. SEO did not become less important when AI platforms entered the picture. It became the prerequisite for showing up in more places.

The Nashville audit captured what four platforms did on one date in one metro for one practice area. The next audit will cover a different market and a different specialty, and the patterns will shift. That shift is itself a finding worth tracking.

Most firms have no mechanism for measuring this. They track Google rankings. Maybe they monitor a few branded search terms. But almost none of them know what ChatGPT says when a prospective client asks for a recommendation. Almost none know whether Perplexity cites their website or a third-party directory when their practice area comes up. Almost none know which phrasings trigger an AI Overview and which do not.

The discovery journey increasingly starts on platforms that traditional analytics cannot track. A prospective client asks ChatGPT for a recommendation, hears a firm's name for the first time, then searches that firm's name on Google. The firm sees a branded search conversion and attributes it to SEO. The actual attribution chain started on a platform the firm is not monitoring, cited a source the firm may not even know exists, and produced a recommendation the firm had no hand in shaping.

The firms that treat AI visibility as an ongoing measurement problem rather than a one-time optimization are building something that compounds. They are tracking which platforms name them, which sources those platforms cite, how phrasing changes the outcome, and how those patterns shift over time. That body of evidence about their own visibility is itself a form of competitive advantage. The firms that do not track it are flying blind in a landscape where three different systems are making three different decisions about who deserves to be recommended.

Search for your own practice area on all four platforms today. Screenshot the results. If your firm's name is missing from two of them, you are standing where 68% of the Nashville firms stood this week. Visible in one place. Invisible everywhere else.

The full breakdown, with all 175 observations, is in the first issue of How Machines Decide.