How it works

Trocial’s methodology

Trocial follows a five-stage loop for every brand it scans: Discover, Measure, Diagnose, Act, Retest. This page explains what actually happens at each stage, using the same fields and evidence that appear in a real AI Recommendation Profile - not a marketing simplification of it.

Check My AI Presence See the methodology

1. Discover

Trocial crawls the submitted website and extracts brand facts - category, core offer, target customer, geography - along with a confidence level for each fact. Facts a model could not confirm are flagged as missing rather than guessed.

2. Measure

Trocial generates realistic buyer-intent prompts and runs them against configured AI providers (OpenAI, Anthropic, Gemini, and Groq, where connected). Each result records the provider, whether it completed successfully, and the brand’s position in the answer: recommended, mentioned, absent, unclear, or negative. A provider that is not configured for a scan is labeled unavailable, never simulated.

3. Diagnose

From the completed checks, Trocial builds a competitor recommendation map showing which named competitors appear in the same buyer-intent answers, how often, and - where the evidence supports it - why a model may prefer them. This is combined with the technical and entity readiness check to separate content gaps from technical (crawlability, schema, metadata) gaps.

4. Act

Findings are converted into a prioritized action plan: a specific finding, the evidence behind it, the recommended change, and an estimated priority and effort. Some actions can generate a draft asset directly - for example, an llms.txt file, FAQ schema, or a buyer-intent page outline - for a person to review before publishing.

5. Retest

Paid plans can schedule the same buyer-intent prompts to re-run on a recurring basis, so a team can see whether recommendation share, visibility, and citation-related signals moved after a fix shipped - a comparison against a real baseline, not a fresh, unrelated score.

The honesty rules behind this

Three rules apply across every scan: a provider that was not actually called is never reported as checked; a metric that cannot be computed from real evidence is shown as pending or not enough evidence, never fabricated; and nothing is auto-published to a customer’s site - every generated draft is reviewed by a person first.

Frequently asked questions