loading workspace…
What actually happens inside a 24-hour Cognoverge audit, end-to-end: how the prompt library is assembled, how engines are probed, and how the composite score lands.
Every Cognoverge engagement begins with a 24-hour audit. Free, no card, no sales call until the firm asks. The audit is a diagnostic, not a sales tool — it tells the firm where its AI visibility actually stands across the eight engines that matter for legal and audit buying decisions.
This is what happens between the form submission and the verdict landing in the inbox.
When a firm submits the audit form, three pieces of metadata enter the system: firm name, primary practice area, and jurisdiction. From these, the Auditor agent constructs the firm's tag set — practice area mapped onto our internal taxonomy, jurisdiction mapped onto regulator overlays, firm size estimated from public registers.
Triage runs a quick eligibility check. We do not run audits for firms whose name conflicts with a current client engagement (the conflicts-wall enforcement), and we deprioritize audits for firms whose practice area is outside our current rulebook coverage.
The library is 200 prompts, constructed from three sources:
The 200 prompts cluster into three intent tiers: high-intent ("which firm should I hire for X"), mid-intent ("describe firms that handle X"), and low-intent ("what is X"). Roughly 40% high, 45% mid, 15% low.
The Sweeper agent issues 200 prompts across 8 engines: ChatGPT (GPT-4-class and GPT-5-class), Claude (latest production), Perplexity (default and focus modes), Gemini (1.5 Pro and Flash), Copilot (Bing-backed), Grok, Meta AI, and Google AI Overviews. That is 1,600 (prompt × engine) probes per audit.
Each probe is a structured request, not a screen-scrape. We use first-party APIs where available (ChatGPT, Claude, Perplexity, Gemini), and respect rate limits and robots.txt for the rest. The verbatim answer is captured per (engine, prompt, timestamp), with citations and source links extracted.
Sentiment is classified by a small classifier we own — better than relying on the engine's own self-described sentiment, which is biased toward the engine's training.
The Scorer computes the firm's composite VIS·IDX on a 0–100 scale. The math is straightforward; the calibration is what took us 18 months to get right.
Per (engine, prompt) we compute:
These aggregate up to per-engine VIS·IDX components, which then weight into the composite. Engines are not weighted equally; ChatGPT and Perplexity together carry roughly 55% of the weight in our default model, based on buyer-intent attribution from our partner data.
Before the audit deliverable is generated, the linter runs a pre-pass on the proposed recommendations. We do not ship recommendations that violate the rulebook for the firm's jurisdiction and firm type.
Typical lint findings on the recommendations layer: comparative language flagged for substantiation, performance claims flagged for COBS 4.6 review (UK), independence-adjacent claims flagged for §1.600 review (audit firms). Roughly 20% of generated recommendations get rewritten or dropped at this stage.
The deliverable is a PDF and a workspace link. The PDF contains:
The workspace link gives the firm read-only access to the live sweep data — same probes, same scoring, refreshed weekly.
The deliverable goes to the email address on the form, with a same-thread invitation to schedule a 20-minute walk-through. We say no to roughly 40% of post-audit walk-through invitations because conflicts, fit, or rulebook coverage do not align — same standard as any boutique consulting relationship.
The audit does not crawl your own website, does not analyze your social presence, and does not compute SEO metrics. We consider those outside the AEO/GEO scope; there are competent SEO tools and we are not one. The audit measures the AI engines' description of your firm — that is the single diagnostic we own.
Run the audit at cognoverge.com/audit. Reference IDs start with A1-; verbatim transcripts within 24 hours.
The free 24-hour audit shows you specifically how the eight engines describe your firm against 200 high-intent legal and compliance queries.