How to Monitor Brand Mentions in ChatGPT and Perplexity

If you want to know whether AI tools recommend your brand, track the same prompts every week in both ChatGPT and Perplexity. I’d watch three numbers first: brand mention rate, citation rate, and AI share of voice. That matters because brands can appear in answers without getting linked, and cited sources can shift by 40%-60% month to month.

Start with a fixed prompt list:

• I’d build a fixed prompt list with discovery, comparison, and decision questions
• I’d run each prompt 3-5 times per platform
• I’d use a new chat/session every time
• I’d save the full answer text, not just yes/no mention data
• I’d log mentions, placement, citations, and competitors
• I’d review ChatGPT and Perplexity separately

A simple starting mix is:

• 40% discovery
• 40% comparison
• 20% decision

And a simple starting size is:

• 20-30 prompts for smaller teams
• 100+ prompts for larger brands with more products or audiences

One point stood out to me: ChatGPT mentions brands more often than it links to them. So if I only look at citations, I’ll miss a big part of brand visibility.

I’d treat this like a weekly scorecard. Run the same prompt set, save the raw outputs, log each answer as its own row, and watch for drops in mention rate, lost source slots, or competitor takeovers on high-buying-intent prompts.

That’s the core process this guide lays out.

AI Brand Mention Tracker (AEO/GEO) in Google Sheets

Build a Prompt Set You Can Track Every Week

Use the same prompts every week. That way, shifts in visibility reflect actual change, not small wording differences. Think of this list as your fixed test set. Build it first, then run it the same way every week.

Pick prompts by discovery, comparison, and decision intent

Map your prompts to how people shop. In most cases, a buyer starts broad, compares a few options, and then narrows in on what fits their situation. Your prompt set should follow that path with three prompt types:

• Discovery prompts ask broad category questions without brand names. For example, "What are the best project management tools for small teams?" Don’t include your own brand here. If you do, you’re priming the model, and that can skew the result.
• Comparison prompts put brands next to each other. For example, "your brand vs. a competitor" These show how the AI positions your brand against other options.
• Decision prompts get more specific about the buyer, team, or use case. For example, "Best CRM for a 10-person sales team." These tend to sit closest to an actual buying choice.

A solid starting mix is 40% discovery, 40% comparison, and 20% decision prompts.

Lock the list and tag each prompt

A good starting range for smaller brands is 20 to 30 prompts. Larger companies with several product lines or buyer types often go to 100 or more. The main thing is simple: pick a set you can run every week without changing it.

After you lock the list, tag each prompt so you can sort and review the data later. At a minimum, tag by engine, funnel stage, intent type, audience, and result type.

These tags make the data much easier to read. You can sort by engine, intent, or mention status and spot where your brand shows up - and where competitors show up instead.

Once the prompt list is fixed, run every prompt in a fresh session and save the raw answers.

Run Clean Tests in ChatGPT and Perplexity and Save the Raw Answers

Use fresh sessions and run each prompt multiple times

Run each prompt the same way every week so your results stay comparable. One-off tests can be misleading because AI answers can change from one run to the next.

A simple rule works well: run each prompt 3 to 5 times per platform and report mention rate, not just a yes-or-no result. Use incognito or private browser windows or stay fully logged out. Then start a fresh chat thread for every prompt so earlier context doesn't leak into the next answer.

Settings matter just as much as the prompt itself. In Perplexity, use a web-searching mode such as Best, Pro Search, Reasoning Search, or Research so the output reflects normal search behavior. In ChatGPT, turn on Search to browse the web. If you skip that step, you're not testing live retrieval. You're testing older training data instead.

For each run, log the date in YYYY-MM-DD format. And save every answer exactly as shown. That gives you a clean record you can check later without guessing what changed.

Save answer text, brand placement, and all visible sources

Save the full response text, not just whether your brand showed up. The wording matters. The tone matters. The framing matters. A brand mention can sound like a strong recommendation, a passing reference, or even a negative comparison.

You should also track where the brand appears in the answer and which competing brands appear alongside it. Placement matters because first-place mentions carry a lot more weight than lower placements.

Perplexity is simpler to review because each answer includes numbered sources, so keep those links exactly as shown. ChatGPT mentions brands 3.2 times more often than it actually links to them, which means unlinked mentions can slip past you if you only scan citations. Read the full answer text. When Browse is on and citations appear, save those URLs too.

It also helps to note one extra detail: whether each mention includes a source link or appears without one. That split between a mention with a source link and a mention without a source link becomes important once you start building reports.

Then log each run in a spreadsheet or AskWatch so you can compare week-to-week results.

Log AI Answers in a Spreadsheet or AskWatch

Set up a row-by-answer tracking sheet

After you’ve saved the raw answers, turn them into data you can track. The main rule is simple: log each answer as its own row, not each prompt.

If you run one prompt 5 times, that should create 5 rows. That’s what lets you measure rates instead of ending up with a simple yes-or-no tally.

Here are the fields your sheet should include:

With that setup, you can compare results by prompt, engine, and competitor without having to reread every single answer. It also keeps the sheet tied to the three metrics that matter here: mention rate, citation rate, and share of voice.

A manual sheet is usually enough when you’re working with fewer than 20-30 prompts.

Use AskWatch to automate prompt monitoring and reporting

Once manual logging starts to drag, move the same field structure into AskWatch.

AskWatch runs your saved prompt set across ChatGPT, Perplexity, and other supported AI engines. It checks both the answer body and the citation list, so brand mentions without links don’t get missed. It also automates weekly prompt runs, saves answer text and cited sources, and turns that data into mention and citation reports you can use right away.

AI citation sources shift month to month. Weekly automated checks help you spot those changes while they’re happening, not weeks later.

Compare Trends by Engine and Turn the Data Into Reports

Review mention rate, citation rate, and competitor substitution

Once your weekly runs are in the sheet, turn that data into trend reports. The first step is simple: separate results by engine before you compare anything.

ChatGPT and Perplexity don’t surface brands the same way. If you lump everything together, you can miss patterns that matter.

Track average position and share of voice to see how visible your brand is when it appears. Those numbers help you spot where competitors are gaining ground and taking attention away from you.

You should also flag any prompt where a competitor:

• replaces you in a top result
• takes your cited source slot
• pushes your mention rate down by more than 20% week over week

Then group prompts by intent stage: awareness, consideration, and decision. That makes it easier to see where you’re losing ground.

A competitor winning decision-stage prompts like "best [category] for [use case]" is a bigger problem than losing an awareness-stage prompt. Why? Because it usually points to a content gap or an authority gap on a specific page.

A repeatable workflow for AI brand mention monitoring

From there, the reporting cadence is pretty simple. Review results weekly or every other week, keep the same prompt set, and use AskWatch to turn raw answers into trends you can report on.

AI citation sources shift 40-60% month to month, so consistency matters. Manual tracking in a spreadsheet works for a smaller prompt set across a few surfaces. Once you go past that, AskWatch can handle the full process by running your saved prompts across ChatGPT and Perplexity, pulling answer text and cited sources, and turning them into weekly or monthly reports you can share with a manager or client.

"A brand can have excellent social sentiment and near-zero AI search presence. Monitoring one does not substitute for monitoring the other." - Sal Morton, Content Marketing Lead, Pulsar Platform

The point is simple: show up when buyers ask AI, and know when a competitor shows up instead.

FAQs

How do I choose the right prompts to track?

Build a fixed library of 10 to 50 conversational prompts that reflect how potential customers look into your category. Skip keyword-planner lists. Write prompts the way a real buyer would ask them in everyday language.

Your prompt set should cover the full buyer journey, including:

• Informational prompts for early research
• Commercial prompts for evaluating options
• Comparative prompts for side-by-side checks
• Navigational prompts for finding a specific brand, product, or page

Don’t focus only on branded terms. Those often show awareness, not discovery. And use the same prompt set for every run so you can compare results over time without muddying the picture.

Why should I measure mentions and citations separately?

Measure mentions and citations separately because they tell you different things.

A mention means your brand made it into the consideration set. A citation links to your domain, which helps with trust and can send direct traffic.

Keeping them separate makes diagnosis much easier. If mentions are missing, you likely have a branding issue. If people mention you but don’t cite you, the problem usually sits with content structure or authority.

How long should I monitor before judging trends?

Because AI responses can change from one prompt to the next, it's smarter to watch longer-term patterns instead of reacting to daily swings. Most experts suggest tracking for at least one full quarter so you can build a steady baseline and figure out which queries matter most.

After that, switch to a monthly cadence for regular monitoring. If you're dealing with high-risk categories or an active product launch, weekly spot checks can help you catch fast changes.