Methodology

Transparency as our golden rule

You deserve to know exactly how Pulsari measures your visibility. Here are the 6 exact steps, no black box, with the limitations and choices we own.

01

Enter your business

30 sec

Name, city, industry, website URL (optional) and a short description. Pulsari builds your profile in 30 seconds.

02

Pulsari generates 15 realistic prompts

10 sec

Our system uses a reference LLM (Claude Sonnet) to generate 15 questions phrased like a real user would: recommendations, comparisons, urgent needs, direct searches.

03

Each prompt is sent to target AIs

45 sec

ChatGPT, Gemini and Claude. Each response is captured raw, timestamped, and stored in your database.

04

Fuzzy mention detection

automatic

A dedicated algorithm detects your brand mention in each response: exact name, domain match, partial keyword match on significant words, with protection against false positives (no substring match on less than 5 characters).

05

Competitor extraction

automatic

For each response, we extract the names of competing businesses, brands or shops mentioned by the AI, with local/national/international classification. Exclusion filter on 30+ platforms (Google Maps, Yellow Pages, LinkedIn...).

06

Visibility score + recommendations

instant

For each engine, the presence rate is the number of tested prompts where the brand is detected, divided by the 15-prompt panel. The global score is the average of these rates. The report includes the responses, detected competitors and actions tied to observed evidence.

Our limitations, which we own

  • · Generated prompts, not collected. Our 15 prompts are produced by a reference LLM based on your industry and city. They are realistic but don't reproduce the full spectrum of real searches in your market.
  • · The score is a sample. 15 prompts give a snapshot, not an absolute truth. A Pulsari report identifies trends, not a market share percentage to the decimal point.
  • · AIs evolve. Models and their responses change over time. Two audits can therefore produce different scores without a change to your site. Compare identical panels and retain the raw responses.
  • · The score measures binary presence. It does not weight mention order, sentiment or the quality of a cited source. Review those dimensions in the report's raw responses.

Our sources and technical stack

AIs queried

Claude Sonnet 4 (Anthropic), GPT-4o mini (OpenAI) and Gemini Flash (Google). Each AI is queried with the same prompt structure for comparability. Perplexity and Google AI Overviews are not included in the current audit.

Technical stack

Supabase for the database, Deno functions for orchestration, React 19 for the interface and Cloudflare Pages for hosting. The queried models are operated by Anthropic, OpenAI and Google; processing terms are described in our privacy policy.

Detection algorithm

Exact name matching, fuzzy on significant words (>= 4 letters), domain match, with word boundary to avoid common false positives.

Platform filter

Over 30 platforms and directories excluded from detected competitors (Google Maps, Yellow Pages, LinkedIn, Trustpilot, etc.).

Test the method on your brand

First audit free. Processing time depends on the queried engines.

Launch my free audit