ChatGPT
Gemini
Google AI Overviews
Peec AI
PostHogAn employment-law firm was famous for one case and invisible for everything else. Its search traffic was news traffic: people following a celebrity lawsuit, not workers looking for a lawyer. In three months, a rebuilt site and a verified content program flipped that. The number one page on the firm's site is an article we shipped in early August, and the articles as a group out-pull the famous case more than 10 to 1 on Google impressions in the first 25 days of September. Here is the work, with the numbers and the caveats.
The client
Delshad Legal is a plaintiff-side employment law firm in Los Angeles: wrongful termination, severance, harassment, retaliation, wage theft. Founder Jonathan Delshad represents workers against employers, and the firm's best-known matter, Dixon v. Tyler Perry, put it in national headlines.
The problem
Headlines are not clients. The firm's search presence was almost entirely its famous case: the lawsuit pages drew tens of thousands of impressions from people following the story, while the searches that actually produce clients, such as "what disqualifies you from unemployment in California," "how much severance is normal," and "can I sue my employer for emotional distress," went to legal directories and bigger firms instead. The site itself was a legacy WordPress build with the rankings to match.
The strategy
Three moves, in order.
Rebuild the foundation. We moved the firm onto our stack with a clean article and practice-area architecture, migrated the URLs, and cleaned up the structure the old build left behind.
Ship verified answers to the questions clients ask. Legal content has a bar most agency content does not: a wrong claim is not embarrassing, it is dangerous. Every article runs through a verification pass where each legal claim is checked against the statute, the case, or the agency source before publish. On that foundation we shipped a battery of articles from late June through September targeting real client questions: unemployment disqualification, final-paycheck law, wage theft, settlement averages, arbitration agreements, emotional distress claims, workplace retaliation, leave and disability.
Point it at AI search. The same questions are now asked to ChatGPT and Google's AI answers, so the pages are structured for engines to lift clean answers, and we track 75 buyer prompts in Peec AI to measure when the firm starts being the cited answer.
The results
Sixteen weeks in, the content beats the headlines 10 to 1
First 25 days of September 2026, Google Search Console, as each group's share of the two combined. "Articles we shipped" are the pages we published since the engagement began on 4 June; the celebrity-case pages are the six pages about the firm's headline matters, together with their pre-migration URLs.
| Page group | Share of clicks | Share of impressions |
|---|---|---|
| Articles we shipped | 87% | 91% |
| Celebrity-case pages | 13% | 9% |
The articles answering client questions now out-pull the famous lawsuit more than 10 to 1 on impressions and 6 to 1 on clicks. In the first seventeen days of August the same split was roughly 3 to 1. That is the flip that matters: news traffic reads and leaves. People searching "what disqualifies you from unemployment" have a problem the firm solves.
The number one page on the site is seven weeks old
The unemployment-disqualification guide went live on 5 August. Seven weeks later it is still the page with the most impressions on the site, at position 5.6 over 1–25 September, with 5.7× the impressions of the celebrity-case article. Around it, the severance guide, the maternity-leave guide, the workplace-retaliation guide and the short-term-disability guide all sit between positions 4 and 7.
The site-wide numbers moved with it
Clicks for June to August are 2.5× the clicks for March to May. August was the firm's best complete month in Search Console's records for both clicks and impressions, and September passed it on both by the 21st: the first 25 days brought 143% of August's impressions and 142% of its clicks, with the average position across the site moving from 15.2 in August to 8.3. Impressions per day are 6.8× the March average.
Enquiries and AI referrals: the traffic reached the firm
Two things a worker can do on the site are measured in PostHog: submit the case-review form, and arrive from an AI assistant. Case enquiries through the form ran 2.7× per week from the start of August to late September against the weeks before August. Visits that arrived from an AI assistant itself (ChatGPT, Gemini, Claude, Copilot, Perplexity) were 6× June's in August, and ChatGPT alone sent nearly ten times as many. The base is single digits in June and tens of visits a month by August, so we report the multiple and not a trend line. The enquiry counts stay private; the chart on this page shows the weekly shape, indexed to the pre-August average.
Update, 28 September 2026: AI search stopped being early, and stayed there
When this page first went up, it said the AI side was "where this program is heading, not where it started." It arrived in August and has held. Over the 30 days to 27 September, Peec AI puts Delshad Legal's AI share of voice at 32.4%, first of the 12 tracked employment-law firms and ahead of the next firm by more than ten points, at an average mention rank of 1.4. Week by week, the share of voice went from 0.13% in the week of 8 June to 33.78% in the week of 21 September, and the firm has ranked first every week since 10 August and above 30% every complete week since 24 August.
| Window | AI share of voice | Rank among 12 tracked firms | Google clicks/week, outside the celebrity pages |
|---|---|---|---|
| Week of 8 Jun 2026 | 0.13% | 11th | baseline (week of 7 Jun) |
| Week of 24 Aug 2026 | 32.71% | 1st of 12 | 6.1× (week of 23 Aug) |
| 30 days to 27 Sep 2026 | 32.4% | 1st of 12 | 7.6× (week of 13 Sep) |
Broken down by engine over the same 30 days: ChatGPT 34.8% share of voice, Google AI Overview 32.3%, Gemini 29.7%, with a mention rank between 1.1 and 2.0 on each. Share of voice is the firm's slice of the mentions that name any of the 12 tracked firms; most answers to these prompts name none of them, so the firm's visibility, the share of all tracked answers that name it, sits between 2.9% and 4.6% by engine. Two caveats on the June baseline: the prompt set was rebuilt in late July and early August, when 51 of today's 75 prompts were added, so the jump in the week of 3 August coincides with the new prompts, and the June readings came from the earlier set, of which one prompt survives. A further 12 prompts were added on 8 September, inside the 30-day window, and one of them was replaced on 24 September. The tracked panel has run 1,500 to 1,700 AI answers a week since late August.
Peec weeks start on a Monday and Search Console weeks on a Sunday, so the Google column names its own week. The week of 23 August read 5.8× when this page first quoted it on 24 August and has read 6.1× since Search Console finalised it. Google clicks per week outside the celebrity-case pages held between 5.6× and 6.1× the week of 7 June from late August to the week of 6 September, then climbed to 7.6× in the week of 13 September; after three flat weeks, clicks are rising with impressions again.
What made this different
Two disciplines. The first is verification: legal content that survives a claims check against primary sources is rare, and it is exactly what both Google and AI engines reward in a field where being wrong has consequences. The second is honesty about what counts: we separated the news demand from the client demand in every report, because a traffic chart inflated by a celebrity lawsuit would flatter us and mislead the client. The numbers above exclude the famous case on purpose. The growth is the part the firm keeps.
Updated 28 September 2026 with Search Console data to 25 September, Peec AI daily and weekly readings and its 30-day window to 27 September, and PostHog to 27 September, with enquiries counted in complete weeks to 26 September.
Engagement: site rebuild plus verified SEO and AEO content program, 2026.
