What Should I Baseline Before Starting GEO Work?
Before starting GEO work, capture five baselines: your AI share of voice across target prompts, your current citation footprint (which URLs get cited where), AI crawler activity in server logs, AI referral traffic in analytics, and competitor positions on the same prompt set. Without this snapshot you cannot prove the campaign moved anything.
Why a baseline decides whether GEO looks like a success
AI answers change every run, models update quarterly, and referral traffic is noisy. If you start optimizing without a recorded starting point, any later improvement is unattributable — you cannot separate your work from a model update or seasonal demand. A defensible baseline, captured over at least two weeks so daily variance averages out, is the difference between "we think GEO worked" and "share of voice rose from 12% to 27% on our 40 core prompts."
The five baselines to capture
Record each of these before you publish or change a single page:
- Share of voice — for a fixed prompt set, what percentage of answers mention your brand across ChatGPT, Perplexity, Gemini, and the other engines you care about. This is your headline metric.
- Citation footprint — which specific URLs get cited, on which engines, for which prompts. Deep-URL detail matters here; knowing "our pricing page is cited in Perplexity" is actionable, "our domain appears sometimes" is not.
- Crawler activity — GPTBot, ClaudeBot, PerplexityBot, and Google-Extended hits in your server logs. If AI crawlers cannot reach your key pages, no amount of content work will help until access is fixed.
- AI referral traffic — sessions from chatgpt.com, perplexity.ai, and similar, isolated in analytics as a pre-campaign volume.
- Competitor positions — the same share-of-voice and citation measures for two or three rivals, so you can track relative movement, not just your own.
How to record it cleanly
| Baseline | Source | Cadence |
|---|---|---|
| Share of voice | Prompt sampling across engines | Daily for 2 weeks |
| Citation footprint | Cited-URL extraction per prompt | Weekly |
| Crawler hits | Server / CDN logs | Continuous |
| Referral traffic | GA4 or server analytics | Continuous |
| Competitor SOV | Same prompt set, tracked rivals | Weekly |
A platform like Menra records these as a starting snapshot through its visibility monitoring so you have a timestamped "day zero" to measure against. If you baseline manually, freeze a spreadsheet with dated screenshots — you will want the receipts when a stakeholder asks whether the campaign worked.
Related questions
Decide what counts as a good share of voice for your category before you set targets, and confirm you can see AI crawlers in your logs so the crawler baseline is real and not assumed.
Frequently asked questions
- How long should the baseline period be?
- Run at least two weeks of prompt sampling before you change anything, because AI answers vary run-to-run. A two-week window smooths daily noise into a stable starting number you can defend later.
- Do I need a tool to baseline?
- You can start manually by logging answers to 10-20 core prompts in a spreadsheet. Tooling becomes worth it once you track dozens of prompts across multiple engines, where manual sampling breaks down.
Keep exploring
See how AI engines talk about your brand — track mentions across ChatGPT, Perplexity, Claude, Gemini and 5 more. Start with Menra