You don't need a $89/month tool to track whether ChatGPT, Perplexity, Claude, or Google AI Mode mentions your brand. You need a spreadsheet, a list of buyer prompts, and 30 minutes once a week. Most B2B brands I work with run this manual protocol for 3-6 months before upgrading to a paid tool — and the manual version captures roughly 70% of the value paid tools claim to deliver.
This article walks through the exact protocol, step by step. By the end you'll have a runnable system you can start using today, a spreadsheet template you can copy, and a clear sense of what "good" and "bad" results actually look like. Free, transparent, and built specifically for teams who don't yet need the automation a paid tool provides.
If after reading this you decide you do need automation, my decision tree for AI search visibility tools covers when to escalate. But start here. Most teams discover they don't need the upgrade as fast as they thought.
What AI Citation Tracking Is
AI citation tracking is the practice of measuring how often AI engines (ChatGPT, Perplexity, Claude, Google AI Mode, Microsoft Copilot) cite your content or mention your brand when answering buyer-relevant questions. A "citation" specifically means the AI engine attributes information to your domain — typically with a clickable link or domain reference. A "mention" is when the AI names your brand without linking to your site. Both matter, but citations are the stronger signal because they drive actual referral traffic.
The distinction matters because teams routinely conflate the two and report inflated AI visibility numbers. An AI saying "tools like [your brand]..." is a mention, not a citation. A response with a clickable link to your domain is a citation. Track them separately.
Manual AI citation tracking takes about 30 minutes per week and produces roughly 70% of the value of a paid tool. The protocol: pick 10-20 buyer-intent prompts, run each across 4 AI platforms (ChatGPT, Perplexity, Claude, Google AI Mode), log whether your brand is mentioned and whether your site is cited, and track the pattern weekly. The bottleneck is your time, not the data. For most SMBs through to mid-market B2B teams, manual tracking is enough — paid tools only become necessary when you outgrow the 20-prompt limit or need daily frequency.
What You'll Need Before You Start
Three things, all free:
A spreadsheet. Google Sheets, Excel, Airtable — any tool that handles rows and columns. I prefer Google Sheets because it's free and shareable. The template structure is below.
Access to four AI platforms. Free accounts on ChatGPT, Perplexity, and Claude. Google AI Mode requires no account (just search a question in Google and the AI Overview appears at the top).
30 minutes of focused time per week. Block this on your calendar — Monday morning works well for most teams. Manual tracking falls apart when it becomes "I'll get to it next week."
That's the whole stack. No subscriptions, no API keys, no integrations. Now the protocol.
Step 1: Build Your Prompt Library
[EXP — At Emvigo the most common mistake I see in first-time manual tracking is teams using their brand name as the prompt. "What is Emvigo" doesn't tell you anything useful — of course an AI engine knows your brand exists. The prompts that matter are category-level prompts your buyers actually type, not brand searches you wish you'd win. Build the prompt library carefully and it becomes the foundation for everything else.]
Build a list of 10-20 prompts in three categories:
Category prompts (40% of your list): Broad questions that buyers in your category would ask. "Best CRM for service businesses." "How do I choose a project management tool?" "What's the difference between Asana and Monday?"
Comparison/alternatives prompts (40% of your list): Questions involving your direct competitors. "Alternatives to [Competitor X]." "[Competitor Y] vs alternatives." "Best [your category] for [your specific persona]."
Use case prompts (20% of your list): Specific scenarios your buyers face. "How to track time across multiple client projects." "Tool for managing a 12-person remote team." "How to send proposals to enterprise clients."
Avoid: Brand-name queries about you specifically ("what is [your brand]"), generic informational queries ("what is software"), or queries you don't realistically expect to rank for. The prompts you choose should be the ones your buyers actually use during research — typically surfaced from sales call recordings, support tickets, or customer interviews.
Once you've built the list, freeze it. Don't change prompts month to month — you need consistent prompts to track trends over time. New prompts can be added quarterly. Old ones should rarely be removed.
Step 2: Set Up Your Tracking Spreadsheet
Create columns for:
This gives you longitudinal data — you can see whether citation rate is trending up, down, or flat for the same prompts over time. The spreadsheet itself is the analysis tool.
Pro tip: Set up a separate tab for each platform so you can filter by platform easily. The four AI platforms behave very differently and you'll want platform-specific trend views.
Step 3: Run the Prompts Across Four Platforms
Run each prompt in each of these four places. The order doesn't matter, but be consistent week to week.
ChatGPT
Use ChatGPT with web search enabled (sometimes called "Browse with Bing" or "ChatGPT Search" depending on which interface). ChatGPT's parametric mode (default, no web search) draws from training data and has higher hallucination rates between 18-55% — it's not a reliable signal for current citation tracking. Always enable web search.
Note: ChatGPT links to external sources in about 31% of web-search responses. When it does link, those are real citations. When it just mentions a brand without a link, that's a mention.
Perplexity
Perplexity is the easiest platform to track because every answer includes numbered citations. If your brand appears, you'll see exactly which URL was cited. Perplexity is also RAG-native — every answer requires citations, so the signal is clean. Run the same prompts here that you ran in ChatGPT.
Perplexity is the easiest AI platform to track manually because every answer includes visible numbered citations. ChatGPT (with web search enabled) links externally about 31% of the time, Google AI Mode cites multiple sources per answer, and Claude does not include URL citations at all. The four platforms behave very differently — only 11% of websites are cited by both ChatGPT and Perplexity, so platform-specific tracking is non-negotiable.
Claude
Claude does not include URL citations in its responses. This makes tracking citations literally impossible — but you can still track brand mentions. Log whether Claude names your brand and which competitors it names. Don't worry about the "Cited?" column for Claude responses; just track mentions.
Google AI Mode
Search your prompt in Google (logged out, incognito). The AI Overview appears at the top of the results when triggered. AI Overviews cite multiple sources per answer, usually 3-6 visible citation links. Log whether you appear in the list.
Note: AI Overviews only trigger on certain query types — they appear on 99.9% of informational queries, 57.9% of question queries, and 21% of all keywords overall. If the AI Overview doesn't trigger for your prompt, log "AIO didn't trigger" and move on. That's information too.
Step 4: Log Your Data, Then Calculate Citation Rate
After running all 10-20 prompts across all four platforms, you'll have between 40 and 80 data points for the week. Now calculate three metrics:
Citation rate per platform. For each platform, divide the number of prompts where you were cited by the total number of prompts. If you appeared in 4 out of 10 ChatGPT responses with web search, your ChatGPT citation rate is 40%.
Mention rate per platform. Same math, but counting mentions instead of citations. Mention rate is always higher than citation rate.
Share of voice. For each prompt, count how many times your brand was mentioned versus all competitors. If you appeared once and three competitors appeared, your share of voice for that prompt is 25%.
These three numbers form your weekly snapshot. Add them to a summary tab in your spreadsheet so you can see the trend over time.
Step 5: Track the Pattern Across 90 Days Minimum
Single-week AI citation data is noise, not signal. AI engines produce different answers for the same prompt across sessions, weeks, and updates — sometimes citing you, sometimes not, on the same query. Reliable AI visibility data requires a 90-day rolling window at minimum. Look at trends, not snapshots. A prompt where your citation rate moves from 20% to 60% over three months is meaningful. The same prompt fluctuating from 30% to 50% to 35% week-to-week is normal variance.
Set a calendar reminder for the same time every week. Run the protocol. After 90 days, you'll have enough data to see real patterns:
- Prompts where citation rate is steadily rising — your AEO work is paying off for these queries
- Prompts where competitors are consistently cited and you aren't — content gaps that need addressing
- Prompts where citations fluctuate widely — the AI is uncertain about who to cite; your content might be close but not crossing the threshold
- Prompts that no longer trigger AIOs or AI responses — Google or another engine may have changed how those queries are handled
The trend matters. Single-week panic is almost always wrong.
What Good Tracking Data Looks Like
Three example scenarios from real client work:
Scenario A: Healthy. Citation rate climbing on 6 out of 10 prompts over 90 days. Share of voice in the 20-40% range for category prompts. Mention rate higher than citation rate (expected). Competitor mentions stable or declining. This is what working AEO looks like.
Scenario B: Plateau. Citation rate flat for 6+ months. Mention rate is reasonable but not improving. Few new pages getting cited. Almost always indicates a content velocity problem — the team is maintaining existing content but not producing new AEO-targeted pieces. Fix: increase publishing of comparison, integration, and use-case content.
Scenario C: Reverse. Citation rate falling. Competitor mentions rising on the same prompts. Almost always indicates that competitors are publishing new AEO content faster than you are, or that they've started review-collection programs that are paying off. Fix: audit competitor recent content and review profiles; identify and close the gap.
[EXP — In my work at Emvigo the most common pattern I see in first-time trackers is Scenario A masquerading as Scenario C. Teams panic at week-by-week dips, assume their AEO work isn't working, and start over with a new strategy. Then in month 4 the pattern reveals itself as actually positive — they were just looking at weekly noise. Patience plus 90-day windows is the cure.]
When to Upgrade to a Paid Tool
Manual tracking is sufficient until one of three things becomes true:
You need more than 20 prompts. If your category has 50+ buyer-relevant query variations and you genuinely want to track all of them, manual breaks down. Paid tools handle hundreds of prompts efficiently.
You need daily frequency. Manual tracking is weekly. For fast-moving markets where citation patterns shift in days, weekly is too slow. Paid tools (Otterly, Peec AI, Profound) check daily.
You need to slice data by region, language, or persona. Manual tracking gives you one view: your team running prompts from one location, in one language. Paid tools simulate prompts from multiple geographies, multiple languages, multiple personas simultaneously. If your business is multi-market, this matters.
If none of those three are true yet, stay manual. The decision tree in our AI visibility tools guide covers when the math changes.
Common Mistakes to Avoid
Five errors I see repeatedly:
One: Using brand-name queries. "What is [your brand]" tells you nothing useful. Use category and competitor queries instead.
Two: Changing the prompt list weekly. You can't track trends if the prompts keep changing. Freeze the list. Only update quarterly.
Three: Running prompts on inconsistent ChatGPT settings. Default mode versus web-search mode produces wildly different results. Pick one (web search enabled) and stick with it.
Four: Conflating mentions with citations. An AI naming your brand isn't the same as citing your site. Track them in separate columns.
Five: Reacting to single-week movements. Weekly variance is normal. Look at 90-day trends. Single bad weeks are usually noise.
Summary
To track AI citations manually: build a 10-20 prompt library covering category, comparison, and use-case queries; run each prompt across ChatGPT (web search enabled), Perplexity, Claude, and Google AI Mode weekly; log mentions, citations, position, and competitor mentions in a spreadsheet; calculate citation rate, mention rate, and share of voice per platform; track the trend across a 90-day rolling window; only upgrade to a paid tool when you outgrow 20 prompts, need daily frequency, or need multi-region tracking. The protocol takes 30 minutes a week and captures 70% of paid tool value.
The biggest mistake teams make with AI citation tracking isn't the protocol — it's quitting before 90 days of data accumulates. Citation patterns take months to compound and reveal themselves. The teams that stick with weekly tracking for a full quarter discover that their work either is or isn't producing results, with enough data to act decisively either way.
Run this for 90 days. Then decide whether to upgrade to a paid tool — or whether the manual version is enough.
Want a fully-built AI citation tracking system — including your prompt library, spreadsheet template, and the first month of data analysed?
We set up AEO measurement systems for B2B brands serious about being found in AI search. Includes your custom prompt library, your tracking spreadsheet, the first 90-day baseline, and a quarterly review process your team can run independently afterward.
FAQ
How long does manual AI citation tracking take per week? About 30 minutes for 10-20 prompts across 4 platforms. Setup takes longer (1-2 hours the first time to build the prompt library and spreadsheet). After that, ongoing weekly tracking is a 30-minute task.
Which AI platform is easiest to track manually? Perplexity, because every answer includes visible numbered citations. ChatGPT (with web search) links externally about 31% of the time. Google AI Mode cites multiple sources visibly. Claude does not include URL citations, so you can only track brand mentions there, not citations.
How often do I need to check my AI citation rate? Weekly is the right cadence for most teams. Daily produces too much noise — AI engines genuinely produce different answers across sessions, so single-day data isn't actionable. Monthly is too slow to catch problems early. Weekly for 90 days gives you the cleanest signal.
What's a good citation rate for AI search? Highly category-dependent. For B2B SaaS with active AEO work, 30-50% citation rate on category prompts over 90 days is strong. 60%+ usually indicates category leadership. Below 15% indicates a real gap that needs addressing. But the absolute number matters less than the trend — climbing is good, flat or falling is the signal to act.
Do I need a separate tool to track Perplexity vs ChatGPT? For manual tracking, no — you can use the same spreadsheet across all platforms. For automated tracking, yes — only 11% of websites are cited by both ChatGPT and Perplexity, so platform-specific tracking is essential. If you upgrade to a paid tool, ensure it queries all relevant engines separately.
Can I track AI citations using Google Search Console? No. GSC tracks Google search results only. AI citations from ChatGPT, Perplexity, and Claude happen on those platforms and are invisible to GSC. GSC does show some Google AI Mode and AI Overview traffic, but that's a small slice of the total AI search picture.
