This study measures exact URL and registrable-domain overlap. It does not measure Google AI Overviews or establish answer accuracy, source authority, rankings, traffic, leads, matters or patient outcomes.
Rotgar Research
Google and AI Source Overlap Across 8 Legal Niches
A descriptive US benchmark of explicit ChatGPT and Gemini citations against paired Google organic Top 20 results across 8 legal niches.
How to read this study
A source-overlap benchmark, not a performance claim
The original four niches and 16-niche extension were collected in different windows and through live versus Standard queue endpoints. Cross-niche differences are descriptive.
Google, ChatGPT and Gemini observations came from DataForSEO surfaces, not repeated logged-in consumer interface sessions.
Executive Summary
- Exact URL coverage. Across 8 included niches, vertical-level ChatGPT estimates ranged from 4.1% to 16.2% and Gemini estimates ranged from 45.0% to 57.6%. Exact estimates, 95% intervals, and denominators are in the chart and evidence table. These are primary task-macro estimates conditional on an eligible citation-bearing AI cell and a complete paired Google Top 20.
- Registrable-domain coverage. Across the same niches, ChatGPT registrable-domain estimates ranged from 18.5% to 48.6% and Gemini estimates ranged from 53.0% to 73.1%. Domain matching uses the analyzer's registrable-domain key rather than a simple host comparison. The two surfaces have different eligible task sets, so side-by-side values are not a paired-platform effect estimate. On common support across the included verticals, URL coverage was ChatGPT 10.7% versus Gemini 50.9% (Gemini minus ChatGPT +40.2 pp; n=248 tasks), and domain coverage was 36.3% versus 66.0% (difference +29.7 pp; n=248). These common-support differences are descriptive.
- ChatGPT–Gemini citation-domain similarity. Across the included niches, mean task-level domain Jaccard ranged from 2.9% to 18.7%; exact niche denominators are shown with the chart.
- Eligibility and missingness. Of 576 expected AI task cells, 515 were eligible for the strict primary URL metric and 21 had no eligible citations. Separately, among 288 paired Google task captures, 267 met the strict Top 20 rule, 12 were successful short pools, and 9 were technical errors. AI citation status and Google status can overlap and must not be summed as mutually exclusive categories.
Prior work and this study's scope
An Ahrefs study published August 11, 2025 examined 15,000 long-tail prompts and reported that, on average, 12% of links cited by ChatGPT, Gemini, and Copilot appeared in Google's Top 10 for the same prompt. This report is an independent niche-specific derivative using US-English questions, Google organic Top 20, explicit ChatGPT and Gemini citations, and the vertical scope named above. The study populations, surfaces, rank cutoffs, and aggregation rules differ, so the percentages are not directly comparable.
Exact URL coverage
The chart places the two primary conditional task-macro estimates on a shared scale, but each surface uses its own eligible task-cell set. The bars must not be subtracted as a paired-platform effect. It does not treat missing cells as zero. Exact eligible counts, 95% task-clustered intervals, pooled secondary estimates, and the available-pool sensitivity appear in the evidence table.
Exact URL coverage by vertical and AI surface
Surface-specific eligible sets; 95% task-clustered intervals and denominators are in the evidence table.
Immigration
Personal Injury
Family Law
Employment Law
Estate Planning & Probate
Criminal Defense
Tax Law
Business Law for Individuals & SMBs
Verified Rotgar Research artifact · DataForSEO surfaces ·
Registrable-domain coverage
This view repeats the surface-specific display after URLs are collapsed to registrable domains. Eligible support can differ between ChatGPT and Gemini, so the side-by-side bars are not a paired-platform contrast. It describes source-domain coincidence only; it is not a source-authority or quality score.
Registrable-domain coverage by vertical and AI surface
Surface-specific eligible sets; 95% task-clustered intervals and denominators are in the evidence table.
Immigration
Personal Injury
Family Law
Employment Law
Estate Planning & Probate
Criminal Defense
Tax Law
Business Law for Individuals & SMBs
Verified Rotgar Research artifact · DataForSEO surfaces ·
Citation-set similarity across AI surfaces
The mean task-level registrable-domain Jaccard was Immigration 7.8% (33/36 eligible tasks); Personal Injury 2.9% (32/36 eligible tasks); Family Law 8.9% (34/36 eligible tasks); Employment Law 6.2% (34/36 eligible tasks); Estate Planning & Probate 5.7% (35/36 eligible tasks); Criminal Defense 3.5% (33/36 eligible tasks); Tax Law 18.7% (35/36 eligible tasks); Business Law for Individuals & SMBs 6.5% (31/36 eligible tasks). The denominator is each task's union of ChatGPT and Gemini citation domains; tasks without an eligible pair are NA.
ChatGPT–Gemini citation-domain Jaccard by vertical
Mean task-level intersection divided by union among eligible paired AI citation sets.
- Immigration7.8%33 of 36 tasks eligible
- Personal Injury2.9%32 of 36 tasks eligible
- Family Law8.9%34 of 36 tasks eligible
- Employment Law6.2%34 of 36 tasks eligible
- Estate Planning & Probate5.7%35 of 36 tasks eligible
- Criminal Defense3.5%33 of 36 tasks eligible
- Tax Law18.7%35 of 36 tasks eligible
- Business Law for Individuals & SMBs6.5%31 of 36 tasks eligible
Verified Rotgar Research artifact · DataForSEO surfaces ·
Primary metric eligibility
Across the report's 576 expected AI task cells, 515 were eligible for the strict URL metric. The normalized bars retain every analyzer status so the displayed overlap estimates are read with their actual denominators.
Primary metric status by vertical and AI surface
Each bar accounts for all expected task cells in its vertical–surface group.
Immigration — ChatGPT
- Eligible24 of 36
- No citations1 of 36
- Google <205 of 36
- Google error6 of 36
Immigration — Gemini
- Eligible24 of 36
- No citations2 of 36
- Google <205 of 36
- Google error5 of 36
Personal Injury — ChatGPT
- Eligible26 of 36
- No citations1 of 36
- Google <206 of 36
- Google error3 of 36
Personal Injury — Gemini
- Eligible23 of 36
- No citations3 of 36
- Google <207 of 36
- Google error3 of 36
Family Law — ChatGPT
- Eligible36 of 36
- No citations0 of 36
- Google <200 of 36
- Google error0 of 36
Family Law — Gemini
- Eligible34 of 36
- No citations2 of 36
- Google <200 of 36
- Google error0 of 36
Employment Law — ChatGPT
- Eligible36 of 36
- No citations0 of 36
- Google <200 of 36
- Google error0 of 36
Employment Law — Gemini
- Eligible34 of 36
- No citations2 of 36
- Google <200 of 36
- Google error0 of 36
Estate Planning & Probate — ChatGPT
- Eligible36 of 36
- No citations0 of 36
- Google <200 of 36
- Google error0 of 36
Estate Planning & Probate — Gemini
- Eligible35 of 36
- No citations1 of 36
- Google <200 of 36
- Google error0 of 36
Criminal Defense — ChatGPT
- Eligible36 of 36
- No citations0 of 36
- Google <200 of 36
- Google error0 of 36
Criminal Defense — Gemini
- Eligible33 of 36
- No citations3 of 36
- Google <200 of 36
- Google error0 of 36
Tax Law — ChatGPT
- Eligible36 of 36
- No citations0 of 36
- Google <200 of 36
- Google error0 of 36
Tax Law — Gemini
- Eligible35 of 36
- No citations1 of 36
- Google <200 of 36
- Google error0 of 36
Business Law for Individuals & SMBs — ChatGPT
- Eligible36 of 36
- No citations0 of 36
- Google <200 of 36
- Google error0 of 36
Business Law for Individuals & SMBs — Gemini
- Eligible31 of 36
- No citations5 of 36
- Google <200 of 36
- Google error0 of 36
Verified Rotgar Research artifact · DataForSEO surfaces ·
Common-support platform comparison
A direct descriptive surface comparison must use the same eligible tasks. Across the included verticals, the common-support set contains 248 tasks for URL coverage and 248 tasks for domain coverage. The table shows each surface's macro mean and Gemini minus ChatGPT in percentage points; no causal or paired-bootstrap claim is attached to this derived display.
Common-support ChatGPT and Gemini comparison
Both surfaces are averaged over the same eligible tasks; differences are descriptive point estimates.
| Scope | URL common n | ChatGPT URL | Gemini URL | Gemini − ChatGPT URL | Domain common n | ChatGPT domain | Gemini domain | Gemini − ChatGPT domain |
|---|---|---|---|---|---|---|---|---|
| Immigration | 23 | 16.9% | 53.3% | +36.4 pp | 23 | 32.3% | 73.7% | +41.4 pp |
| Personal Injury | 23 | 4.6% | 56.9% | +52.3 pp | 23 | 18.8% | 68.8% | +50.0 pp |
| Family Law | 34 | 11.0% | 48.8% | +37.8 pp | 34 | 39.4% | 66.9% | +27.5 pp |
| Employment Law | 34 | 9.8% | 52.4% | +42.5 pp | 34 | 40.1% | 70.0% | +29.9 pp |
| Estate Planning & Probate | 35 | 10.1% | 57.6% | +47.5 pp | 35 | 35.0% | 68.1% | +33.1 pp |
| Criminal Defense | 33 | 13.4% | 48.4% | +35.0 pp | 33 | 37.0% | 57.9% | +20.9 pp |
| Tax Law | 35 | 13.9% | 47.1% | +33.3 pp | 35 | 48.1% | 71.2% | +23.1 pp |
| Business Law for Individuals & SMBs | 31 | 5.6% | 45.0% | +39.4 pp | 31 | 31.9% | 53.0% | +21.1 pp |
| Legal report scope | 248 | 10.7% | 50.9% | +40.2 pp | 248 | 36.3% | 66.0% | +29.7 pp |
Evidence table
Primary macro estimates give each eligible task cell equal weight. Pooled estimates instead weight cells by cited-source count and are secondary. Available-pool sensitivity uses fewer than 20 Google URLs where necessary and must not be read as the strict primary estimate.
Coverage estimates and denominators
Primary task-macro, secondary pooled, and available-pool sensitivity estimates are shown separately.
| Vertical | AI surface | URL macro [95% CI] | URL pooled (secondary) | URL available-pool sensitivity | Domain macro [95% CI] | Domain pooled (secondary) | Domain available-pool sensitivity | Primary eligible cells | Available-pool eligible cells | Expected cells | No citation | Google pool <20 | Google unavailable | Other primary status |
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| Immigration | ChatGPT | 16.2% [10.3%, 22.3%] | 17.6% | 13.4% | 30.9% [21.1%, 41.0%] | 27.9% | 28.6% | 24 | 29 | 36 | 1 | 5 | 6 | 0 |
| Immigration | Gemini | 53.6% [41.9%, 65.4%] | 52.2% | 53.9% | 73.1% [62.2%, 83.8%] | 68.7% | 70.9% | 24 | 29 | 36 | 2 | 5 | 6 | 0 |
| Personal Injury | ChatGPT | 4.1% [0.9%, 8.0%] | 6.1% | 4.1% | 18.5% [8.6%, 30.4%] | 16.9% | 19.0% | 26 | 32 | 36 | 1 | 7 | 3 | 0 |
| Personal Injury | Gemini | 56.9% [46.7%, 67.4%] | 55.0% | 50.8% | 68.8% [59.3%, 78.2%] | 66.0% | 62.2% | 23 | 30 | 36 | 3 | 7 | 3 | 0 |
| Family Law | ChatGPT | 11.6% [6.6%, 17.2%] | 11.4% | 11.6% | 41.4% [30.3%, 52.9%] | 29.9% | 41.4% | 36 | 36 | 36 | 0 | 0 | 0 | 0 |
| Family Law | Gemini | 48.8% [38.8%, 59.1%] | 51.5% | 48.8% | 66.9% [56.5%, 76.8%] | 63.4% | 66.9% | 34 | 34 | 36 | 2 | 0 | 0 | 0 |
| Employment Law | ChatGPT | 9.3% [5.2%, 13.8%] | 9.7% | 9.3% | 38.6% [29.0%, 48.8%] | 33.1% | 38.6% | 36 | 36 | 36 | 0 | 0 | 0 | 0 |
| Employment Law | Gemini | 52.4% [42.5%, 62.4%] | 50.4% | 52.4% | 70.0% [60.6%, 79.0%] | 69.1% | 70.0% | 34 | 34 | 36 | 2 | 0 | 0 | 0 |
| Estate Planning & Probate | ChatGPT | 10.2% [5.4%, 16.2%] | 9.5% | 10.2% | 36.8% [27.2%, 47.4%] | 32.3% | 36.8% | 36 | 36 | 36 | 0 | 0 | 0 | 0 |
| Estate Planning & Probate | Gemini | 57.6% [49.5%, 65.6%] | 58.9% | 57.6% | 68.1% [59.9%, 75.8%] | 68.6% | 68.1% | 35 | 35 | 36 | 1 | 0 | 0 | 0 |
| Criminal Defense | ChatGPT | 14.0% [9.1%, 19.2%] | 15.1% | 14.0% | 39.1% [28.3%, 50.4%] | 32.7% | 39.1% | 36 | 36 | 36 | 0 | 0 | 0 | 0 |
| Criminal Defense | Gemini | 48.4% [38.4%, 58.4%] | 52.4% | 48.4% | 57.9% [46.6%, 68.5%] | 62.4% | 57.9% | 33 | 33 | 36 | 3 | 0 | 0 | 0 |
| Tax Law | ChatGPT | 14.1% [9.4%, 19.2%] | 12.9% | 14.1% | 48.6% [37.3%, 60.2%] | 36.0% | 48.6% | 36 | 36 | 36 | 0 | 0 | 0 | 0 |
| Tax Law | Gemini | 47.1% [36.3%, 57.9%] | 48.4% | 47.1% | 71.2% [60.2%, 81.8%] | 69.3% | 71.2% | 35 | 35 | 36 | 1 | 0 | 0 | 0 |
| Business Law for Individuals & SMBs | ChatGPT | 5.6% [3.0%, 8.5%] | 5.5% | 5.6% | 32.9% [23.1%, 43.4%] | 25.4% | 32.9% | 36 | 36 | 36 | 0 | 0 | 0 | 0 |
| Business Law for Individuals & SMBs | Gemini | 45.0% [35.4%, 54.5%] | 46.4% | 45.0% | 53.0% [44.0%, 62.0%] | 53.2% | 53.0% | 31 | 31 | 36 | 5 | 0 | 0 | 0 |
Frequently observed citation domains
The following audited tables rank registrable domains only by observed frequency in this report's task set. Counts and distinct-task support are descriptive; they are not authority, quality, trust, or optimization scores. Example source URLs are retained in the tables for auditability and are not used to make narrative claims.
Most frequently observed AI-cited registrable domains
Top 10 per AI surface by distinct task count; frequency is not authority or quality.
Most frequently observed matched registrable domains
Top 10 per AI surface among domain matches; frequency is not authority or quality.
Data quality
The final analyzer summary records 2160 of 2160 expected surface rows and no unexpected rows. Strict Google Top 20 completeness was 680 of 720 tasks across the full loaded panel and 267 of 288 tasks in this report. Across the full panel, the other 40 Google tasks comprise 20 successful captures with fewer than 20 unique eligible URLs and 20 technical-error captures; within this report those counts are 12 and 9. Incompleteness by included vertical is Immigration 11/36; Personal Injury 10/36; Family Law 0/36; Employment Law 0/36; Estate Planning & Probate 0/36; Criminal Defense 0/36; Tax Law 0/36; Business Law for Individuals & SMBs 0/36. A short-pool check with denominator 700 describes successful Google captures only; overall noncompleteness uses all 720 tasks. The report gate rejects any non-passing error-severity check.
Analyzer checks not marked pass
Warnings and informational checks are retained; no non-passing error-severity check is allowed by the report gate.
| Check | Scope | Severity | Status | Affected | Total | Affected rate | Details |
|---|---|---|---|---|---|---|---|
google_short_success_below_top_n | task_platform_counts | warning | warn | 20 | 720 | 2.8% | available_unique_counts={"IMM-010":9,"IMM-016":8,"IMM-017":8,"IMM-021":8,"IMM-028":9,"MED-DENT-002":9,"MED-DENT-016":8,"MED-DENT-024":9,"MED-DENT-025":9,"MED-FERT-005":9,"MED-FERT-008":9,"MED-FERT-025":18,"MED-FERT-034":9,"PI-007":19,"PI-009":9,"PI-012":9,"PI-018":19,"PI-019":9,"PI-020":9,"PI-024":9} |
google_strict_unique_top_n_complete | task_platform_counts | warning | warn | 40 | 720 | 5.6% | status_counts={"complete":680,"short_success":20,"technical_error":20};short_available_unique_counts={"IMM-010":9,"IMM-016":8,"IMM-017":8,"IMM-021":8,"IMM-028":9,"MED-DENT-002":9,"MED-DENT-016":8,"MED-DENT-024":9,"MED-DENT-025":9,"MED-FERT-005":9,"MED-FERT-008":9,"MED-FERT-025":18,"MED-FERT-034":9,"PI-007":19,"PI-009":9,"PI-012":9,"PI-018":19,"PI-019":9,"PI-020":9,"PI-024":9} |
google_technical_error_runs | task_platform_counts | warning | warn | 20 | 720 | 2.8% | task_ids=["IMM-015","IMM-018","IMM-019","IMM-020","IMM-022","IMM-025","MED-DENT-006","MED-DENT-008","MED-DENT-010","MED-DENT-011","MED-DENT-015","MED-DENT-018","MED-DENT-033","MED-DENT-034","MED-DENT-036","MED-FERT-012","MED-FERT-031","PI-003","PI-034","PI-035"] |
run_technical_errors | runs | warning | warn | 20 | 2,160 | 0.9% | by_platform={"google":20} |
source_metadata_missingness | sources | warning | warn | 11 | 57,360 | 0.0% | missing={"domain":0,"title":11}; metadata are preserved evidence but not metric keys |
normalized_duplicate_rows_removed | normalized_sources | info | info | 3,006 | 23,134 | 13.0% | deduplication grain=(task_id,platform,normalized_url) |
Methodology
20-niche exploratory source-overlap panel uses a frozen 720-question manifest; this report contains 288 questions across 8 named niches, with 36 questions per niche. Each question pairs Google, ChatGPT, and Gemini captures. Google eligibility uses organic results sorted by group rank, normalized-URL deduplication, and the first 20 unique eligible URLs. AI inputs are explicitly cited sources only.
The primary URL and domain metrics divide the within-task matched AI source count by that AI cell's unique cited-source count, then take an unweighted macro mean across eligible cells. A citation-bearing eligible cell with no match is zero. A no-citation cell is NA and is counted separately. Registrable domains use the analyzer's pinned Public Suffix List logic. The pooled estimate is secondary; the available-pool estimate is a labeled sensitivity for present Google pools below the strict Top 20. ChatGPT–Gemini domain Jaccard is computed within task before the vertical macro mean. No source-class taxonomy is inferred. Primary coverage intervals use 10,000 percentile bootstrap replicates with seed 20260807, resampling tasks within vertical and retaining each sampled task's available surface cells.
All 720 manifest rows retain reviewer_status=unreviewed. No licensed medical or legal domain reviewer evaluated the question set or answers. Therefore this study supports source-overlap findings only, not clinical/legal accuracy, safety, or completeness.
The dataset was collected through DataForSEO mixed live and Standard queue endpoints, not through repeated logged-in consumer UI sessions. Endpoint observations were chatgpt: /v3/ai_optimization/chat_gpt/llm_scraper/live/advanced, /v3/ai_optimization/chat_gpt/llm_scraper/task_get/advanced; gemini: /v3/ai_optimization/gemini/llm_scraper/live/advanced, /v3/ai_optimization/gemini/llm_scraper/task_get/advanced; google: /v3/serp/google/organic/live/advanced, /v3/serp/google/organic/task_get/advanced. Returned collection-time model labels were chatgpt: gpt-5-5, gpt-5-6; gemini: 3.5 Flash-Lite; google: not_exposed. In this report, ChatGPT and Gemini mean those mediated surfaces and returned labels; endpoint behavior may differ from consumer UI.
The 16 added niches were selected after aggregate results for four original niches had been inspected. Original and extension observations were collected in different windows and through live versus Standard queue endpoints; vertical is therefore confounded with timing and retrieval mode. Cross-niche results are descriptive.
Scope and reproducibility
Immigration: topic anchors — employment-based immigration, family-based green cards, and naturalization; intent families — Authoritative-source seeking, Comparison/choice, Eligibility/requirements, Local cost, Local provider discovery, and Process/timeline
Personal Injury: topic anchors — car-accident claims, slip-and-fall claims, and wrongful-death claims; intent families — Authoritative-source seeking, Comparison/choice, Eligibility/requirements, Local cost, Local provider discovery, and Process/timeline
Family Law: topic anchors — adoption, child custody, and divorce; intent families — Authoritative-source seeking, Comparison/choice, Eligibility/requirements, Local cost, Local provider discovery, and Process/timeline
Employment Law: topic anchors — wage-and-hour claims, workplace discrimination claims, and wrongful termination claims; intent families — Authoritative-source seeking, Comparison/choice, Eligibility/requirements, Local cost, Local provider discovery, and Process/timeline
Estate Planning & Probate: topic anchors — probate administration, will contests, and wills and trusts; intent families — Authoritative-source seeking, Comparison/choice, Eligibility/requirements, Local cost, Local provider discovery, and Process/timeline
Criminal Defense: topic anchors — DUI charges, felony charges, and misdemeanor charges; intent families — Authoritative-source seeking, Comparison/choice, Eligibility/requirements, Local cost, Local provider discovery, and Process/timeline
Tax Law: topic anchors — IRS audits, criminal tax investigations, and federal tax debt and collection; intent families — Authoritative-source seeking, Comparison/choice, Eligibility/requirements, Local cost, Local provider discovery, and Process/timeline
Business Law for Individuals & SMBs: topic anchors — business formation, buying or selling a small business, and commercial contract disputes; intent families — Authoritative-source seeking, Comparison/choice, Eligibility/requirements, Local cost, Local provider discovery, and Process/timeline
The full exact wording, geography, topic anchor, and intent family for all 720 questions remain in FULL-QUERY-MANIFEST.csv. The frozen estimands and normalization rules are in PROTOCOL-EXTENSION.md; collection and analysis departures are recorded in DEVIATIONS.csv. The artifact source metadata includes the exact file identities and SHA-256 hashes used for this build.
What this means for Rotgar
Rotgar can use the observed page-level, domain-level, and surface-specific gaps as a measurement baseline for source-discovery and content-distribution hypotheses. The report helps separate Google visibility from AI citation visibility and makes missingness explicit before prioritization. Any optimization experiment should be tracked prospectively; this snapshot alone does not identify why a source appeared or what action would change that outcome.
What this means for law firms
Law firms can use this benchmark to see how source overlap varies across sampled client questions and legal niches. It can guide separate measurement of Google visibility and AI citation visibility, but it does not predict rankings, citations, matters, legal accuracy, or service quality.
Recommended next steps
- Keep this artifact as the immutable 20-niche exploratory source-overlap panel sector view with its input hashes.
- Repeat the same charts in a later, separately frozen observation window before interpreting stability over time.
- Review cited pages behind material descriptive gaps as qualitative follow-up, recording hypotheses separately from measured overlap.
- Track any Rotgar or client experiment against a new, prospectively defined outcome rather than treating Wave 1 overlap as a causal benchmark.
Further questions
- Which page and domain overlaps recur in a later frozen observation window?
- Do observed patterns remain after stratifying by intent family and geography without creating underpowered slices?
- Which cited pages warrant a separate reviewed qualitative taxonomy, and what annotation protocol would make that classification reproducible?
- How sensitive are conclusions to technical missingness and later platform/model changes?
Limitations
All 720 manifest rows retain reviewer_status=unreviewed. No licensed medical or legal domain reviewer evaluated the question set or answers. Therefore this study supports source-overlap findings only, not clinical/legal accuracy, safety, or completeness.
The dataset was collected through DataForSEO mixed live and Standard queue endpoints, not through repeated logged-in consumer UI sessions. Endpoint observations were chatgpt: /v3/ai_optimization/chat_gpt/llm_scraper/live/advanced, /v3/ai_optimization/chat_gpt/llm_scraper/task_get/advanced; gemini: /v3/ai_optimization/gemini/llm_scraper/live/advanced, /v3/ai_optimization/gemini/llm_scraper/task_get/advanced; google: /v3/serp/google/organic/live/advanced, /v3/serp/google/organic/task_get/advanced. Returned collection-time model labels were chatgpt: gpt-5-5, gpt-5-6; gemini: 3.5 Flash-Lite; google: not_exposed. In this report, ChatGPT and Gemini mean those mediated surfaces and returned labels; endpoint behavior may differ from consumer UI.
This is a 20-niche exploratory source-overlap panel snapshot, not a causal study or a performance forecast. Within these 8 niches, 267 of 288 paired Google task pools met the strict unique Top 20 rule; primary estimates exclude the rest, and the evidence table reports the available-pool sensitivity separately. The noncomplete tasks are distributed as Immigration 11/36; Personal Injury 10/36; Family Law 0/36; Employment Law 0/36; Estate Planning & Probate 0/36; Criminal Defense 0/36; Tax Law 0/36; Business Law for Individuals & SMBs 0/36; this report does not assume that missingness is interchangeable across verticals. No-citation AI cells remain NA rather than zero.
The study observes explicitly exposed citations, not every source used internally by a model. URL or domain coincidence does not establish answer accuracy, source quality, endorsement, user attention, rankings, traffic, leads, matters, or patient outcomes. ChatGPT and Gemini surface-specific estimates can use different eligible tasks; their raw difference is not presented as a paired effect. The 95% task-clustered percentile intervals describe sampling variability under the stated bootstrap, not causal uncertainty. Platform models and returned sources can change after capture. Rank-position distributions and source-class claims are outside this report.
The 16 added niches were selected after aggregate results for four original niches had been inspected. Original and extension observations were collected in different windows and through live versus Standard queue endpoints; vertical is therefore confounded with timing and retrieval mode. Cross-niche results are descriptive.
Artifact provenance
Verified inputs remain traceable
The published page is generated from the verified report artifact. The build stops if its receipt, SHA-256, niche set or denominators no longer match the approved evidence package.
Artifact SHA-256cfdcc193e03b5052c4ea373ae8bc021220461d116ea5d93301d22257559d6610
Review evidence identities
- Expanded ICP panel paired task metrics
sha256:9e11fbc71161f3e3b9849f45273c6d83ba528a78c73753b03ec81338258f9933 - Expanded ICP panel aggregate metrics
sha256:26024fd79907e43a5a3cfa6dd63b665d846034395f5b63fa7f538ca4fcc3700e - Expanded ICP panel task-level AI domain Jaccard
sha256:3cfd40580e3659eeb7f332f5598abce611e1efd1ef81ca4a572b0cc942001c37 - Expanded ICP panel data-quality checks
sha256:34d941d5cd2ab029c64936f52d46099ea4667e4515ac9ac203e9ca237032cf79 - Expanded ICP panel deduplicated sources
sha256:09e2764a31ed9a5b2359d3ecb1cc6d6f0d7c4f516225cd9e61b17386e6a1f4fc - Expanded ICP panel paired Google matches
sha256:c76097b6cea95f956446e017ced1c8d9825c9e3828b9b001aa214dd86b1783fa - Expanded ICP panel analysis summary
sha256:8081c495575736e82dc1630dd240282d40a51e712e12082a1dee406772db55ef - AI source-overlap protocol
sha256:79134959f213e7ad67bd430f679b680bebf741fb27dd3904745a62dcf9dfc421 - Frozen 720-question manifest
sha256:bf9692248628180555a7d95d3b5b85b037ebcaf52480c2a88ece46341f920bc0 - Study deviations register
sha256:6439adae3132037cf0c4419298e32612fd126dcead641bbe4b69c891e89d0744 - Expanded ICP panel collection summary
sha256:489a4f089791a63fcb7e6ef6c8a971de30b31d1cc7eec755cf1f0842fdd2a495 - Expanded ICP panel endpoint and model run ledger
sha256:5453efa3591ad002a5f0821e549bb130e40dcf3268bebb6cba6cce63a1c3fead - Deterministic Expanded ICP panel report derivation
sha256-input-bundle:23bb060606221774ce71e9e8c0fbb059a4b9c885538d6d79999435948c1f5658 - Executed Expanded ICP panel report materialization queries
sha256:28bda2b5765623e09cf63b5d9769359747bbeb5b3f0ba9059d78cc3376f306df - Ahrefs AI search overlap study
ahrefs-ai-search-overlap-2025-08-11
Free audit
Start with a free law-firm Google + AI visibility audit
The simplest free audit starts with one law firm or selected office, one priority market and one prospective-client language.
Best fitLaw firms and multi-office legal practices in:
- Personal injury
- Family law
- Criminal defense
- Immigration
- Employment law
- Business + corporate
- Litigation + dispute resolution
- Real estate + property
- Estate planning + probate
- Tax law
- Your firm’s current visibility and named competitors
- The sources AI answers rely on
- A prioritized list of improvements to implement
Several firms, offices, countries, markets or prospective-client languages can be included without a call. Prefer to discuss the scope? Book a 30-minute call We’ll tailor the audit at no cost.