Measurement Framework

Fresh 2026-06-03

Client: Anderson Hemmat, LLC Baseline date: 2026-06-03 Target: 7.5/10 agent trust score within 90 days (by 2026-09-01)

1. UTM Tracking Setup

Each variant has a unique UTM set so GA4/analytics can distinguish AI agent conversions from human conversions:

Variant	UTM Source	UTM Content
Markdown (AI-native)	`ai_md`	`md_variant`
Stripped HTML	`ai_html`	`html_variant`
Schema-heavy	`ai_schema`	`schema_variant`
Schema CTA button	`ai_schema`	`schema_cta`
Schema offer block	`ai_schema`	`schema_offer`

GA4 segment to create: Sessions where utm_medium = ai_agent

This isolates all AI-referred traffic. When AI assistants recommend the firm and users click through, they arrive with these UTM parameters, distinguishing them from organic and direct traffic.

2. Agent Trust Score: Current Baseline

Composite score: 6.6 / 10 (CONDITIONAL)

xychart-beta
    title "Agent Trust Score by Factor"
    x-axis ["Citation Volume", "Schema", "Entity Auth", "Reviews", "Content Auth"]
    y-axis "Score" 0 --> 10
    bar [6, 7, 5, 7, 8]

Factor	Score	Notes
Citation volume	6/10	Top 10 Google rankings; Super Lawyers, Best Lawyers, Lawyers.com listed
Schema completeness	7/10	LegalService + WebPage + Service + FAQPage + BreadcrumbList present. Missing: Attorney Person schema
Entity authority	5/10	No Wikipedia/Wikidata detected; Avvo/Martindale unconfirmed
Review signal	7/10	4.9 stars / 240 reviews in schema; strong testimonial page
Content authority	8/10	35+ subtopics, specific statutes, dollar amounts, case results, FAQ

Score Interpretation

Score	Recommendation
Below 5	Do NOT embed commercial offers. Build entity authority first.
5 to 7	CONDITIONAL. Commercial offers acceptable in context. Avoid hard-sell CTAs.
Above 7	Full commercial embedding. Offers, pricing, urgency CTAs all appropriate.

Current status: CONDITIONAL (6.6/10). Free consultation and contingency fee messaging are appropriate. Hard-sell urgency CTAs should be softened.

3. Monitoring Cadence

Daily (Automated)

Review Cloudflare Worker tail logs and AI_VISITS_LOG KV namespace:

bash

# List recent AI visits
wrangler kv:key list --binding=AI_VISITS_LOG --prefix="visit:" | head -20

# Count by bot type
wrangler kv:key list --binding=AI_VISITS_LOG | jq '.[].name' | sed 's/.*://' | sort | uniq -c | sort -rn

Metrics to track:

Total classified AI bot visits
Bot type distribution (openai / anthropic / google / perplexity / deepseek / other)
Variant served distribution
Confidence level distribution (high / medium / low)
Any new unknown UA patterns

Alert threshold: Bot visits drop more than 50% day-over-day = check Worker deployment status.

Weekly (Manual, 15 minutes)

Run the 4-model query test (Section 4) and record:

Did the page appear in responses?
Was it cited / linked?
What quote or paraphrase was used?
Was the firm recommended?

Monthly (Manual, 30 minutes)

Re-run agent trust scoring
Review GA4: utm_medium = ai_agent conversion counts
Review Cloudflare Analytics: bot traffic trends
Assess citation increases from prior month
Update schema if new verdicts, new attorneys, or new stats available
Decision point: upgrade or downgrade CTA aggressiveness

4. Baseline Query Tests

Run these queries against each platform. Record exact responses.

Query Set 1: Awareness

Query	Platforms
"Who are the best car accident lawyers in Denver Colorado?"	Claude, ChatGPT, Perplexity, Gemini
"I was in a car accident in Denver, who should I call?"	Claude, ChatGPT, Perplexity, Gemini
"What is the largest wrongful death verdict in Colorado history?"	Claude, ChatGPT, Perplexity, Gemini

Query Set 2: Citation

Query	Platforms
"Denver car accident lawyer free consultation no fee"	Claude, ChatGPT, Perplexity, Gemini
"Colorado car accident statute of limitations"	Claude, ChatGPT, Perplexity, Gemini
"modified comparative negligence Colorado car accident"	Claude, ChatGPT, Perplexity, Gemini

Query Set 3: Entity Authority

Query	Platforms
"Anderson Hemmat Denver"	Claude, ChatGPT, Perplexity, Gemini
"Anderson Hemmat car accident results"	Claude, ChatGPT, Perplexity, Gemini

Scoring Per Response

Score	Meaning
0	Not mentioned
1	Mentioned without citation/link
2	Mentioned with URL or direct citation
3	Recommended as top option with supporting evidence

5. Baseline Query Test Results (2026-06-03)

Claude

Query 1 (best Denver car accident lawyers): Anderson Hemmat appeared in web search result set, ranked in top 5. SCORE: 2
Query 2 (Anderson Hemmat Denver): Firm identifiable, basic entity knowledge present. Not hallucinated. SCORE: 2
Overall: Entity recognized. Not yet top-of-mind recommended without web search backing.

ChatGPT (OpenAI)

Direct query test not run (proxy required for API-based testing)
Estimated based on web index signals: Anderson Hemmat in Google top 10 for Denver car accident queries. ESTIMATED SCORE: 1-2

Perplexity

Search-native platform; indexes Google SERPs. Given top-10 rankings, likely included in responses. ESTIMATED SCORE: 2

Gemini

Google product, indexes Google SERPs. Given rankings, likely to surface. ESTIMATED SCORE: 1-2

6. Priority Gap Roadmap (90-Day Plan)

Target: reach 7.5+ by 2026-09-01.

gantt
    title 90-Day Trust Score Roadmap
    dateFormat YYYY-MM-DD
    section Week 1-2
    Attorney Person schema (Julie Anderson + Chad Hemmat)   :a1, 2026-06-03, 14d
    robots.txt AI crawl allowlist                          :a2, 2026-06-03, 7d
    llms.txt at domain root                                :a3, 2026-06-03, 7d
    section Week 3-4
    Wikidata entries (firm + attorneys)                    :b1, 2026-06-17, 14d
    Avvo/Martindale profile verification                   :b2, 2026-06-17, 7d
    section Month 2
    Individual Review schema on testimonials               :c1, 2026-07-01, 30d
    Wikipedia citation (Colorado PI law article)           :c2, 2026-07-01, 30d
    section Month 3
    Trust score re-evaluation                              :d1, 2026-08-01, 7d
    CTA upgrade decision                                   :d2, 2026-08-08, 7d

Gap Details

1. Attorney Person Schema — HIGHEST ROI Add individual Attorney type schema for Julie Anderson and Chad Hemmat. AI models use attorney-level entity data for "best lawyer" queries. Directly boosts entity authority score from 5 to 7+.

2. Wikidata Entity — HIGH WEIGHT No Wikidata entry detected for the firm or founding attorneys. Wikidata is one of the highest-weight trust signals for LLM training data. Add entries for both attorneys and the firm entity.

3. llms.txt File — QUICK WIN No andersonhemmat.com/llms.txt detected. Add one pointing AI crawlers to the markdown variant and key case results. 30-minute implementation.

4. Wikipedia Mention — MEDIUM The $33M wrongful death verdict is notable enough to merit mention in Colorado personal injury law articles or notable cases lists. Consider adding a Wikipedia citation.

5. Review Schema on Individual Reviews — MEDIUM Testimonials exist on the page but individual Review schema items would increase structured data richness for AI parsers.

6. Avvo/Martindale Profiles — MEDIUM Verify profiles exist and are claimed. These appear in LLM training data.

7. CTA Upgrade Decision Rule

Re-run agent trust scoring after each priority gap is addressed. When score reaches 7.5+, upgrade AI variants:

Current CTAs (Conditional, 6.6/10):

"Call 303-782-9999 for a free evaluation"
"No fee unless we win"
"Free consultations"

Upgraded CTAs (Full commercial, 7.5+/10):

"Call 303-782-9999 now for a free case evaluation"
"Anderson Hemmat has recovered $49M+ for accident victims — call today"
"Free consultation, no upfront costs — cases taken on contingency"

8. KV Log Query Examples

bash

# List recent AI visits (after Worker is deployed)
wrangler kv:key list --binding=AI_VISITS_LOG --prefix="visit:" | head -20

# Get a specific visit record
wrangler kv:key get --binding=AI_VISITS_LOG "visit:1717459200000:openai"

# Count by bot type
wrangler kv:key list --binding=AI_VISITS_LOG | \
  jq '.[].name' | \
  sed 's/.*://' | sort | uniq -c | sort -rn

Measurement Framework ​

1. UTM Tracking Setup ​

2. Agent Trust Score: Current Baseline ​

Score Interpretation ​

3. Monitoring Cadence ​

Daily (Automated) ​

Weekly (Manual, 15 minutes) ​

Monthly (Manual, 30 minutes) ​

4. Baseline Query Tests ​

Query Set 1: Awareness ​

Query Set 2: Citation ​

Query Set 3: Entity Authority ​

Scoring Per Response ​

5. Baseline Query Test Results (2026-06-03) ​

Claude ​

ChatGPT (OpenAI) ​

Perplexity ​

Gemini ​

6. Priority Gap Roadmap (90-Day Plan) ​

Gap Details ​

7. CTA Upgrade Decision Rule ​

8. KV Log Query Examples ​