What is a perception audit?
A perception audit queries multiple AI systems about your brand and analyzes their responses. It's like running a focus group, but the participants are ChatGPT, Claude, Gemini, Perplexity, Grok, and Mistral.
The audit reveals what each AI says about your brand, where responses are accurate vs. inaccurate, how you compare to competitors in AI recommendations, and gaps in AI's knowledge about your product.
Running your first audit
Step 1: Set up your brand - Enter your brand name, website, and a brief description. This helps VectorGap understand what accurate responses should look like.
Step 2: Configure audit scope - Choose which AI providers to include: ChatGPT (largest user base), Claude (strong reasoning), Gemini (Google ecosystem), Perplexity (citation-focused).
Step 3: Select query types - Pick the types of questions: brand queries, comparison queries, category queries, feature queries.
Step 4: Run the audit - VectorGap queries each provider and collects responses. This takes 2-5 minutes depending on scope.
Reading your results
Your main score (VPS - VectorGap Perception Score) is a composite of all six metrics, ranging from 0-100.
0-40: Significant issues - AI perception is hurting your business
41-60: Room for improvement - Some gaps to address
61-80: Solid foundation - Minor optimizations needed
81-100: Excellent - AI accurately represents your brand
What to do with results
Don't try to fix everything at once. Prioritize:
1. Critical inaccuracies - Wrong pricing, fake features, entity confusion
2. Competitive positioning - Are you recommended appropriately?
3. Coverage gaps - Key differentiators that aren't mentioned
Before you run the audit
Write down the version of truth you expect AI to repeat: category, ICP, offer, pricing, geography, proof, and competitors. Without that baseline, you will confuse “different wording” with an actual perception problem.
Include prompts from the full buying journey: category discovery, competitor comparison, pricing, risk, implementation, and alternatives.
After the audit
Do not export a report and stop. Convert every finding into one of four actions: update public truth, add missing source evidence, correct stale third-party profiles, or monitor because the answer is acceptable.
Fix public surfaces first. Internal knowledge helps diagnostics, but AI systems need public, crawlable evidence to change what they say.