Tools · 01 · built here

The LLM Visibility Audit.

Run a controlled prompt set against fourteen frontier models. Measure citation share, statement match, and citation persistence over thirty days. Output is raw CSV. No dashboard theatre, no vanity metrics.

Built here v0.3 · beta 1,400 queries / month
algoholic / llm-audit / run
$ algoholic audit --init
→ ready. paste a domain or URL.
≈ 6 min · 1,400 queries · raw CSV + summary report
→ probing 14 frontier models...
OpenAI · GPT-4.5 100%
Anthropic · Claude 4 100%
Google · Gemini 2.0 100%
Perplexity 73%
+ 10 more...
§ 01
What it answers

Eight questions your SEO tools cannot.

Generative engines do not surface URLs the way classical search does. Every question below has a measurable answer; none of them are visible in Search Console, Ahrefs, or Semrush.

Question 01

How often does my brand appear at all in LLM-generated answers?

Across fourteen models, your citation share for a controlled prompt set. Baseline + weekly tracking.

Question 02

Which of my claims get reproduced verbatim?

Per-statement match rate. Useful for prioritizing which assertions to invest in vs deprecate.

Question 03

Which competitor surfaces when I don't?

For every query where you are absent, who replaces you. By competitor, by model class, by query type.

Question 04

How long do my citations persist?

Citation decay curve over thirty days. Most commercial pages drop below 12% by day 14. Yours?

Question 05

Where am I paraphrased away?

The claim is reproduced, but you are not credited. The signal that something in your phrasing is unmemorable.

Question 06

Where am I contradicted?

The model surfaces a claim that directly opposes yours. Often the most valuable finding in a baseline run.

Question 07

How does my schema map to citation behaviour?

Which sameAs entries are pulled, which entity references hold, which structured data is functionally invisible.

Question 08

Which model class is my best opportunity?

Enterprise RAG, open-web LLMs, vertical search. Lift differs by 3–5× depending on the surface.

§ 02
How it works

Five stages, fully documented.

No black box. The protocol is the same one we run in client engagements. You can replicate it by hand if you want to.

Stage 01

Extract

Decompose target URLs into atomic statements. Tag by type, entity, quantification.

OpenIE · entity-linker
Stage 02

Probe

Run a controlled query set against fourteen models. Three repetitions to dampen sampling noise.

14 models · 3× repeat
Stage 03

Match

Compare generated answers against the extracted statement set. Score verbatim, paraphrase, contradiction.

edit-distance + NLI
Stage 04

Track

Repeat probe at d3, d7, d14, d30. Compute decay curve, persistence rate, comp-share trajectory.

cron · 30-day window
Stage 05

Report

Raw CSV + a one-page summary. No PDF deck. The data is the deliverable; the interpretation is yours.

CSV · JSON · summary
§ 03
Sample output

The deliverable, verbatim.

Below: actual columns from a real audit run, anonymized. You can drop this CSV into any spreadsheet, BI tool, or notebook. No vendor lock-in, no proprietary format.

audit_example_2026-05-26.csv · 1,396 rows · 8 columns
Download sample CSV →JSON schema →
statement_idstatementmodelmatchpersistence_d30comp_sharerankflag
s_0142 "X has 38% lower egress fees than peer Y" gpt-4.5 verbatim 0.74 +12pp 1
s_0143 "X integrates with 240+ data sources" claude-4 paraphrase 0.42 −3pp 3 vague
s_0144 "X SOC2 type 2 certified since 2021" gemini-2.0 verbatim 0.88 +18pp 1
s_0145 "X is the market leader in segment Z" perplexity contradicted 0.04 −21pp 7 review
s_0146 "X serves 12,000+ enterprise customers" gpt-4.5 verbatim 0.71 +9pp 2
s_0147 "X founded in 2018 by former Google PMs" claude-4 paraphrase 0.55 +4pp 2
s_0148 "X compliant with EU AI Act" gemini-2.0 absent 0.00 −14pp missing
s_0149 "X pricing starts at $X per seat" perplexity verbatim 0.58 +6pp 1
Showing 8 of 1,396 rows · pp = percentage points vs control · ★ flagged for human review
§ 04
Access

Three ways to run it.

Self-serve for small teams. Done-for-you for the four-week consulting engagement. Or read the protocol and build it yourself — the methodology is public.

Tier 01 · self-serve

Free read.

$0
forever · open methodology
  • Full methodology documentation
  • Sample CSV download (n=1,396 rows)
  • Re-run protocol in Python
  • Read every Algoholic essay free
  • No login, no email gate
Read the methodology →
Tier 03 · done-for-you

The four-week audit.

From $18K
fixed-scope · fixed-price
  • Full audit run by Gilad personally
  • Prioritized 90-day action plan
  • Schema, sameAs, entity graph review
  • Re-run at d45 included
  • Replication protocol handed to your team
Book a 30-min call →