How the AI Clout Score works.

For every domain we scan, we generate 20–40 buyer-intent prompts — the kinds of questions a real customer types into ChatGPT or Perplexity right before buying. Categories include:

• Direct discovery ("best tools for X")
• Comparison ("X vs Y")
• Problem-led ("how do I fix Z")
• Brand-named ("is <brand> any good")

Prompts are stored on your project — you can read every one on the AI Visibility page.

Each prompt is sent to a panel of large language models. Currently in rotation:

• Gemini 2.5 Flash (Google)
• GPT-5 Mini & Nano (OpenAI)
• Optional: Claude (Anthropic) on Multi-Model Comparison

We don't query Perplexity's hosted product directly — instead we model the same buyer-intent prompts on the underlying frontier LLMs, since that's what powers the answer layer of every modern AI search.

Every prompt result lands in one of three buckets:

Your AI Clout Score is a weighted percentage:

score = round(
  (strong × 1.0 + weak × 0.4 + invisible × 0.0)
  / total_prompts × 100
)

• 0–39: Mostly invisible. AI engines are recommending other people.
• 40–69: Showing up sometimes. Inconsistent across engines.
• 70–100: Consistently cited. AI knows what you do and recommends it.

We never call you "Dominating" or "#1". We don't make up tiers we can't defend.

For each prompt we extract every named brand or product in the response and surface them on your Competitors page. The scrape is a real fetch via Firecrawl — not a synthetic guess.

The Action Recommendations engine reads each invisible/weak prompt and produces 4–5 specific fixes (content, schema, backlinks, social, comparison). Effort is estimated in hours so you can prioritize.

• Confidence band. Because LLMs are non-deterministic, your score is shown as a range (e.g. 47 ±4). The midpoint is our best estimate; the band reflects how mixed your prompt results are. A "Stable" badge appears when the band is tight and the sample is large enough to trust.
• The same prompt can produce a slightly different answer twice in a row. We sample at scan time — your score is a snapshot, not gospel.
• Brand-name detection is string-based with normalization. Very generic brand names may inflate counts. You can flag false positives in Settings.
• We don't crawl Google. We don't track keyword rankings. This is an AEO tool, not an SEO tool.
• We never publish your scan. Public report links are opt-in per project.