You don't need a tool to audit your GEO performance. You need a spreadsheet, a few hours, and a disciplined set of prompts. This guide walks you through the exact audit I run for early-stage B2B SaaS, with a scoring system you can re-run every month.
Why bother with a manual audit?
Paid GEO tools are useful for ongoing monitoring, but they don't replace the muscle of actually reading what AI tools say about you. The audit you run yourself shows nuance the dashboards miss - tone, omissions, the company you're listed alongside.
What do I need before I start?
- Free accounts on ChatGPT, Perplexity, Claude and Gemini.
- A spreadsheet (Google Sheets is fine).
- A list of 5 close competitors.
- Roughly 2-3 hours of uninterrupted time.
Step 1: pick your prompts
Use the table below as your starter pack. Adapt each example to your product, category and buyer. Aim for 10-15 prompts across these types.
Step 2: run each prompt in all four tools
Open a fresh chat for each prompt in each tool (no prior context). Copy and paste the prompt exactly as written. Capture the full response in your spreadsheet, along with any cited sources. Take a screenshot if you want a visual record.
Run from a logged-out or incognito state where possible. Personalised context can skew results.
Step 3: score the results
For each prompt in each tool, give yourself a score out of 3 using the rubric below. You'll end up with a grid of, say, 12 prompts × 4 tools = 48 scores. Sum them for a single visibility number.
Step 4: read the failure patterns
The interesting analysis isn't the total score - it's the pattern of zeros and ones. Common patterns and what they tell you:
- Zero on definitions: entity recognition problem. Fix homepage and Organization schema.
- Zero on category recommendations: off-site authority problem. Get on G2, listicles, Reddit.
- Mentioned but inaccurate: conflicting signal between site, schema and third parties.
- Mentioned in Perplexity but not ChatGPT: probably an indexing recency issue - ChatGPT's web index lags.
- Mentioned everywhere except for specific feature questions: product page is unstructured or hidden behind JS.
Step 5: turn it into a backlog
For each failing prompt, write a one-line hypothesis ("we're not in 'best CS tools' answers because we have no G2 reviews") and a one-line action ("get 5 G2 reviews by end of Q3"). You should end with 5-10 specific actions, ordered by leverage.
Step 6: re-run monthly
The audit is only useful if you repeat it. Calendar the same audit, same prompts, every 30 days. The trend matters more than any single score.
For the metrics layer beyond a single audit, see GEO metrics that actually matter: what to track and ignore.