Founder’s Guide to GEO Analytics: Tracking Mentions Across ChatGPT & Perplexity

Introduction

By the end of this guide, you’ll have a repeatable system to track whether ChatGPT and Perplexity mention your brand, how they describe you, and (when citations exist) which URLs they rely on. You’ll also have a baseline score you can improve week over week—so your company becomes the definitive first choice in the new AI economy, not by hope, but by measurement.

Before you start, you need:

• A written list of your brand identifiers (company name, product name, website, founders, social handles).
• Access to ChatGPT (chatgpt.com) and Perplexity (perplexity.ai). Perplexity Threads created while signed out are not saved long-term and can disappear after 14 days, so sign in for reliable tracking.
• A place to log results (a spreadsheet is enough).

Steps

Phase 1: Build a baseline you can trust

1. Define what “a mention” means for your brand.

   Pick one clear rule and write it at the top of your tracking sheet. Use this simple definition:

   • Mention = your brand name OR product name appears in the answer, a list of tools, or a recommendation.
   • Strong mention = you’re recommended (not just referenced).
   • Wrong mention = you’re mentioned but described inaccurately (this is entity drift, and it matters).

   Checkpoint: you can answer “If ChatGPT says X, do I count it?” without debating.

2. Create your canonical “entity profile” in one document.

   This is the grounding document you’ll use whenever an AI system gets you wrong. Create a doc called “Brand Entity Profile” and include:

   • Official company name (exact spelling and capitalization).
   • Common misspellings and name variants (what people actually type).
   • Product names and short descriptions (1 sentence each).
   • Your primary website domain and 2–5 key URLs you want cited (homepage, pricing, docs, “about,” comparison page).
   • Founder names and titles (only if you want them to be part of the public entity).
   • 3 proof points that must be correct (e.g., “We do X,” “We integrate with Y,” “We serve Z”).

   Checkpoint: you have a single source of truth you can paste into prompts or use to correct misinformation.

3. Build a 25-question prompt set that matches how customers ask.

   Open a blank document and create 5 categories with 5 prompts each (25 total). Keep them unbranded first (meaning: you do not mention your company name). This is the real test of AI visibility.

   • Category discovery: “What are the best tools for [job-to-be-done]?”
   • Problem/solution: “How do I [painful task] as a [persona]?”
   • Alternatives: “What are alternatives to [top competitor] for [use case]?”
   • Comparison: “Compare [competitor A] vs [competitor B] for [use case].”
   • Vendor shortlists: “Recommend 5 vendors for [category] and why.”

   Checkpoint: you can copy/paste a prompt set that represents real buying intent, not vanity questions.

4. Create a GEO tracking sheet and add these columns.

   In a spreadsheet (or your preferred tracker), make one row per prompt per platform. Use these columns exactly (you’ll thank yourself later):

   • Date (YYYY-MM-DD)
   • Platform (ChatGPT or Perplexity)
   • Prompt (paste exact text)
   • Response link (shared link URL)
   • Mention (Yes/No)
   • Mention strength (Strong/Weak/Wrong)
   • If mentioned: how described (1 sentence you copy from the answer)
   • Citations present (Yes/No)
   • If cited: cited URLs (paste the list)
   • Accuracy issues (None/Minor/Major)
   • Notes (anything weird: outdated info, confused competitor, hallucinated feature)

   Checkpoint: you can fill one row in under 60 seconds.

5. Lock ChatGPT into “clean slate” mode before you test.

   You want consistency, not personalization. The simplest way is to use Temporary Chat.

   • In ChatGPT, start a new chat and click the pill-shaped “Temporary” button in the top-right corner of the page.
   • Temporary Chats won’t appear in your history, and ChatGPT won’t remember what you discuss in them.

   Optional (recommended if you benchmark often): turn off memory so regular chats don’t affect future tests. Go to your profile picture > Settings > Personalization, then disable memory controls (your exact toggles may vary by plan).

   Checkpoint: you have a testing environment where yesterday’s chats don’t leak into today’s results.

6. Run your prompt set in ChatGPT and save a shared link for each result.

   For each of your 25 prompts:

   • Paste the prompt into your Temporary Chat and submit.
   • After ChatGPT answers, click the Share button on the top-right of the chat (or share from the sidebar).
   • Preview the snapshot, then copy the link and paste it into your tracking sheet.

   Two important details:

   • A shared link is accessible to anyone who has it, and it includes the full conversation up to the point you shared (it’s a snapshot). Do not include sensitive information in benchmarking chats.
   • If you turned off “Chat History & Training,” the Share button is disabled for conversations; in that case, rely on Export instead.

   Alternative capture method (bulk): export your ChatGPT data via Settings > Data Controls > Export Data, then Confirm export, and download the .zip from your email. The export includes your chat history in a file named chat.html.

   Checkpoint: every ChatGPT result in your baseline has a share link or is recoverable via export.

7. Lock Perplexity into a consistent mode before you test.

   Perplexity can behave differently depending on whether Web is on and which Focus mode you choose, and you can’t change certain settings mid-thread. So decide your standard now.

   • Sign in so your Threads are saved long-term.
   • Decide whether you will test with Web ON (recommended for real-world visibility) or Web OFF (useful if you want “model-only” answers). You can toggle Web using the globe icon at the start of a new thread, but only at the beginning.
   • If you use Focus modes, set it at the start of a new Thread; you can’t switch Focus within an existing Thread.

   Checkpoint: you can describe your Perplexity test settings in one sentence (example: “Signed in, Web ON, Focus: All/Internet”).

8. Run the same prompt set in Perplexity and share each Thread link.

   For each of the same 25 prompts:

   • Start a new Thread (do not reuse Threads across unrelated prompts; you want clean comparisons).
   • Paste the prompt and submit.
   • When the answer appears, click Share at the top-right of the Thread and set it to “Sharable,” then copy the link into your tracking sheet.
   • Copy the source URLs Perplexity shows and paste them into your “cited URLs” column (this is how you learn what Perplexity trusts).

   Optional organization (highly recommended): create a dedicated Space called “GEO Tracking” and store your Threads there so your team can collaborate and you can keep projects separated.

   • Click the Spaces icon in the left-side panel, then click Create new Space.
   • Spaces are private by default, and you can share later using the Share button and Copy Link.

   Checkpoint: each Perplexity result has (1) a sharable link and (2) a pasted list of sources.

9. Score your baseline and calculate your 4 GEO KPIs.

   Once you’ve filled 50 rows (25 prompts × 2 platforms), calculate:

   • Mention Rate = (mentions) / (total prompts)
   • Strong Mention Rate = (strong mentions) / (total prompts)
   • Citation Rate (Perplexity) = (prompts where your domain is cited) / (total prompts)
   • Drift Rate = (wrong mentions) / (mentions)

   Don’t overcomplicate it. You’re looking for direction and leverage:

   • Where are you invisible?
   • Where are you visible but misrepresented?
   • Where are you cited, and which pages are being used?

   Checkpoint: you can point to 5 prompts where you’re winning and 5 where you’re missing, with proof links.

Phase 2: Turn baseline data into compounding visibility

10. Mirror your entity inside geOracle.ai’s Perception Engine to monitor drift and citation patterns.

    Manual benchmarking shows you the snapshot. The Perception Engine is how you keep watching without living in tabs. Use it to create a single “digital twin” of your brand entity so you can:

    • Detect entity drift (AI systems confusing your brand, mislabeling features, or mixing you with similarly named companies).
    • Monitor citation patterns (which URLs are showing up, and whether that’s changing).
    • See your AI visibility as an entity graph rather than a pile of screenshots.

    Action to take: in Perception Engine, create (or add) your brand entity and paste the exact identifiers from your Entity Profile (official name, variants, domain, key URLs, core proof points). Save it as your canonical entity.

    Success looks like: the status board shows your entity, its known variants, and a clean baseline “this is us” representation you can monitor.

11. Use the Strategy Roadmap to turn missing mentions into a weekly execution plan.

    This is where founders get stuck: they collect data, then do nothing with it. Don’t do that.

    Action to take: create a Roadmap called “GEO Visibility” with 3 phases:

    • Phase 1: Baseline (complete)
    • Phase 2: Fix foundation (technical + entity clarity)
    • Phase 3: Publish for citations (content that models can retrieve and trust)

    Then, add the 10 worst-performing prompts (no mention, or wrong mention) as your first tasks. That’s your execution queue.

    Success looks like: you have a prioritized list that tells you exactly what to do next week, not “we should improve GEO.”

12. Run a GEO Analyzer audit on the pages you want AI systems to cite.

    If Perplexity is citing other sites, you don’t win by complaining—you win by being easier to retrieve, easier to parse, and easier to trust.

    Action to take: in GEO Analyzer, audit your top 3–5 “citation target” URLs (pricing, docs, comparison, category landing page).

    What you’re looking for:

    • AI readability issues (pages that are technically accessible but hard for retrieval systems to extract facts from).
    • Indexing/crawl blockers (anything that prevents content from being discovered).
    • Thin or ambiguous content (pages that don’t clearly state what you are, who it’s for, and why you’re different).

    Success looks like: you have a short list of concrete fixes tied to specific URLs (not a vague “improve SEO” directive).

13. Use Sitemap View to identify the content gaps that explain your missing mentions.

    Most “no mention” results come from a simple problem: the web doesn’t contain a clear, authoritative page that answers the exact question the user asked.

    Action to take: open Sitemap View and map your current pages against your 25 prompts. For each prompt where you’re missing, ask:

    • Do we have a page that answers this exact question?
    • Is it obvious from headings and the first screen what the answer is?
    • Does it include comparable alternatives and decision criteria (so an AI can confidently recommend it)?

    Success looks like: you can name 3 new pages you need to publish (and why).

14. Publish your first 1–3 “AI-citable” pages, then re-test the exact same prompt set.

    Action to take: publish pages that directly match your highest-intent missing prompts. You can write manually (best for nuance) or use the Article Platform to accelerate the workflow (best for speed and consistency across formats and channels).

    If you use the Article Platform, use it like a founder, not like a content mill:

    • Feed it the exact prompt (as the “search intent”).
    • Paste your Entity Profile proof points so the content cannot drift.
    • Generate the article, then edit the first 20% yourself (headline, intro, key claims). That’s where trust is won.
    • Publish, then let it generate distribution content (social posts) so the page actually gets discovered.

    Now re-run your 25 prompts in both platforms using the same clean-slate settings and update your KPIs.

    Final checkpoint: you can show a “before vs after” change with saved evidence links, and you know which actions caused the lift.

Troubleshooting

Problem: “My results change every time, so tracking feels pointless.”

This is normal. You’re measuring a probabilistic system. What makes it actionable is consistency in your testing setup.

• Use Temporary Chat in ChatGPT so memory doesn’t leak into benchmarks.
• In Perplexity, don’t reuse Threads across prompts, and don’t change Web/Focus settings mid-stream (you can’t). Start a new Thread with the same settings each time.
• When a prompt is critical, run it 3 times and score the most common outcome (not the best outcome).

Problem: “ChatGPT won’t let me share the conversation.”

If you disabled Chat History & Training, ChatGPT disables the share button for conversations.

• Use Export Data as your capture method instead (Settings > Data Controls > Export Data > Confirm export).
• Or temporarily enable chat history for the benchmarking session, then turn it off again (if that fits your policy).

Problem: “Perplexity mentions me but never cites my site.”

That usually means Perplexity can’t find a page that cleanly supports the claim it wants to make, so it cites someone else.

• Look at what Perplexity is citing instead (your sources column) and ask: what do those pages have that yours doesn’t?
• Use GEO Analyzer to improve AI readability and fix indexing blockers for the pages you want cited.
• Publish a direct “answer page” that matches the prompt exactly (one page per high-intent question beats one vague homepage).

Conclusion

When you’re done, you’ll have:

• A tracking sheet with a repeatable prompt set, proof links for every result, and baseline KPIs for ChatGPT and Perplexity.
• A clear view of where you’re invisible, where you’re winning, and where AI systems are getting you wrong.
• A living execution plan (Strategy Roadmap) that turns “we should improve GEO” into weekly tasks you can actually complete.
• A Perception Engine “digital twin” that helps you monitor entity drift and citation patterns—so your brand stays the definitive first choice as AI systems evolve.

That’s what GEO Analytics is for: not vanity charts, but a tight loop of measure → diagnose → publish → verify. If you run this loop weekly, you’re not just reacting to the AI economy—you’re building your company to lead it.