Which Boulder law firms does AI recommend? What auditing ten small firms showed

No firm came out as the one AI recommends. When our June 2026 audit ran "best [practice] lawyer in Boulder" across six practice areas, the top positions went to a few larger local firms set aside as benchmarks, and to the legal directories Justia, Avvo, FindLaw and Super Lawyers. The ten solo and small firms split five weak, five not.

Which firms did the answers put first?

Not the solo and small firms. The batch notes record a few larger local firms and the legal directories holding the top positions for Boulder legal searches. None of the ten graded Dominant, the tier for a business named first or second in effectively every answer its batch ran.

The ten split into two Invisible, three Partial, three Moderate and two Strong. Invisible, Partial and Compromised count as weak, so that is 5 of 10: the law offices row of the Front Range AI Visibility Index, where 33 of 66 owner-led businesses across five industries graded weak. A cell of ten is texture, not a ranking.

We name none of the ten, weak or strong, and none of the benchmark firms. The directories are the only names here. What we do for firms, and in which practices, is on our page on AI search and GEO for Colorado law firms; this article stays with the evidence and the method.

What did the audit ask?

One question shape across six practice areas: "best [practice] lawyer in Boulder", for family law, personal injury, estate planning, criminal defense and DUI, immigration, and real estate. Each firm's scan notes then record the specific phrasings that firm was checked against.

The last row sits outside the six areas in the recorded prompt, but the scan notes carry it.

Why did the phrasing of the question matter?

Because a firm could be named for one description of its work and absent for another. Four of the ten were named for one phrasing of their practice and missing, or clearly weaker, for another. A fifth led the answers for one of the practice areas it offers and trailed for another.

The splits ran along lines a client would recognize: a general practice term against a specific service inside it, such as probate or employment law, and narrow property matters against the broad real-estate search. Two firms surfaced only when someone searched the firm's name. One was named every time its practice was searched, and never first.

The notes do not say why a phrasing missed. Google's guide to its generative AI features says its systems understand synonyms, and that creating separate content for every search variation, primarily to manipulate rankings or AI responses, violates its scaled content abuse policy. That guidance covers Google Search only. But probate and employment law are services, not synonyms. If you do the work, say so on the site and on every listing. A page for every rewording is the move Google warns against.

What patterns ran through the ten scans?

Reputations off Google, profiles the research could not confirm, and sites that undersold the attorneys. The table counts each pattern across the ten scan files, and says so where free-text notes would not support a count.

Reviews. The review bases that sat off Google were on platforms such as Avvo, Birdeye, Yelp and Lawyers.com, and only one scan recorded no weakness in reviews at all. The batch summary says Google counts were routinely not retrievable, uses labeled proxies instead, and lists verifying each firm's live Google rating and count as the first field task for nearly every firm.

Profiles. Of those seven, two were flagged as likely unclaimed. One firm had no profile, one appeared claimed, and one scan did not record the status. Unconfirmed describes what the research could see as much as what the firm has: something to check, not a verdict.

Sites. The credentials were real: peer ratings, teaching roles, decades at the bar. The visible defects were the kind a prospective client notices in seconds: broken contact details, a template error, a site that did not look like an established practice's. We describe them no more closely than that, because a specific defect is as identifying as a name. Firms without online booking relied on a phone call or a contact form. The facts that disagreed across listings were an office address in one case and a founding date in the other.

What can this data not tell you?

Less than a table of counts suggests. Six limits apply to every number above, and to the grades.

• Ten firms, one city. They were chosen as plausible clients, not sampled at random, so this does not describe Colorado lawyers generally. The Boulder page covers the city's other three batches.
• Named, not cited. The scans record which firms were named, not which URLs or sources were cited. The law notes also do not name the engine behind each answer, and several describe search snippets, map results and best-of lists alongside AI summaries.
• Proxy review counts. Most review figures come from platforms other than Google and are labeled as estimates in the batch files.
• No causes. Nothing here shows that fixing a pattern changes a grade. The notes' comments on structured data are left out, because the batch's own next steps call for a structured-data test before telling a firm it has none.
• Personal injury is included. Two of the ten practice personal injury; we do not take that work, but the data includes them.
• Not a repeatable measurement. Grades came out of the NSL Growth Snapshot pipeline. Answers vary from run to run, the prompts were not uniform across the Index's seven batches, and a re-run would not reproduce every row.

What can a Boulder attorney do this week?

Three checks, none of which needs a vendor: run the prompt check on each practice you offer, confirm your Google Business Profile, and make your site and directory listings agree.

1. Run the prompt check. Use the phrasings in the table, plus the words your clients use: "probate" as well as "estate planning". Use a fresh session, run each prompt more than once, and date every row. The method is in how to check if ChatGPT recommends your business.
2. Confirm the Google profile. Claim and verify it, then check the category, address and hours. Google's guidelines for representing your business say a solo attorney at a branded firm is best served by sharing one profile named "[firm]: [attorney name]"; where several public-facing attorneys share a location, the firm keeps its own profile and each attorney's carries only their name; and one practitioner should not hold several profiles for different specializations. More in the Google Business Profile guide.
3. Make your listings agree. Name, address, phone, practice areas and bar admissions, stated the same way on your site, your Google profile, and on Avvo, Justia, FindLaw, Super Lawyers, Lawyers.com and Martindale. The local AI search guide covers keeping one business stated one way.

If you ask clients for reviews, stay within each platform's rules. Google's policy allows asking genuine clients but not offering incentives or asking only the ones you expect to be happy, and Yelp asks businesses not to ask for reviews at all. What Colorado's rules for lawyers permit is a question for your own ethics counsel, and this article does not answer it.

Then run the free scan. It reads the page you give it plus robots.txt, sitemap.xml and llms.txt: 26 scored checks across eight dimensions, in seconds. It does not query ChatGPT, Perplexity or Google, and it cannot see a Google Business Profile, so it sits beside the prompt check rather than replacing it. How the scan grades is published.

Where we apply this

The trades and cities our research covers, and the service behind it.

More in Law firms

See what AI search can read about your firm.

We audited ten solo and small Boulder firms in June 2026. The free scan runs the site-side checks on yours; the conversation is where we go through what a client's question actually returns.

What we do for this industry Run the free scan