How to Track Brand Mentions in Perplexity, ChatGPT, Claude, and Gemini

By the end of this tutorial, you’ll have a repeatable system to:

• Track brand mentions in ChatGPT, Perplexity, Claude, and Gemini

• Capture citations, sentiment, pros and cons, and recommendation frequency

• Log results in a consistent schema your team can report on

• Use those insights to make better ecommerce and GEO/AEO decisions

This guide is a hands-on companion to the deeper strategy piece, AI Recommendation Analytics Pillar: Measuring Brand Mentions Across Major LLMs.

Prerequisites

Before you start, make sure you have:

• Access to all four assistants:

  • ChatGPT with web-enabled Search (or equivalent web-connected mode)

  • Perplexity (logged-in account recommended)

  • Claude with web access enabled

  • Gemini with browsing enabled

• A clear list of entities to track:

  • Brand name(s) (e.g., “BrandX”)

  • Key product lines or categories

  • Top competitors

• Basic tooling:

  • A spreadsheet or database (Google Sheets, Excel, Airtable, or a warehouse)

  • Optional: A BI tool (Looker, Tableau, Power BI) or an AI visibility platform like Era

• Team alignment:

  • One owner responsible for running and maintaining the tracking

  • Agreement on which markets, languages, and product categories to prioritize

1. Define the questions you want AI assistants to answer

The most common failure is starting with random prompts. You end up tracking noise instead of decision-stage recommendations.

Your goal: Design a small set of high-intent questions that reflect how real customers use Perplexity vs ChatGPT vs Claude vs Gemini when deciding what to buy.

1.1 Choose question types

Use three core question types:

1. Category discovery

   • Example: “What are the best running shoes for flat feet?”

   • Purpose: See if your brand appears at all and which brands dominate.

2. Comparison / shortlist

   • Example: “Compare BrandX running shoes with BrandY for marathon training.”

   • Purpose: See how assistants frame pros and cons and which brand they lean toward.

3. Branded decision

   • Example: “Is BrandX a good option for trail running shoes?”

   • Purpose: Measure sentiment and recommendation frequency for your specific brand.

1.2 Capture markets and languages

Decide for each question:

• Primary market: e.g., US, UK, Germany

• Language: e.g., English, German, Spanish

Concrete example:

• Query: “Best espresso machine under $500?”

• Market: US

• Language: English

• Product category: Espresso machines

Common failure: Only testing brand-name queries. That measures awareness, not whether you win competitive, generic queries where most new customers start.

2. Standardize a log structure for AI assistant answers

Without a consistent schema, you can’t compare Perplexity vs ChatGPT vs Claude vs Gemini or automate reporting.

Your goal: Create a single table where every answer from any assistant can be logged the same way.

2.1 Use a unified log schema

Create a sheet or table with at least these columns:

• timestamp (ISO format, e.g., 2026-09-15T14:32:00Z)

• region (e.g., US, DE)

• language (e.g., en, de)

• assistant (chatgpt | perplexity | claude | gemini)

• query_text

• brand_name

• brand_mentioned (Y/N)

• brand_rank (1, 2, 3… or NULL if not listed)

• recommendation_flag (Y/N – did it explicitly recommend your brand?)

• sentiment (positive / neutral / negative)

• pros_listed (short text or comma-separated)

• cons_listed (short text or comma-separated)

• citations_present (Y/N)

• citation_sources (primary domains, e.g., example.com, amazon.com)

• competitors_named (comma-separated list)

• purchase_intent_level (awareness / consideration / purchase-ready)

• notes (hallucinations, anomalies, etc.)

2.2 Example row

Imagine you ask Perplexity: “Best trail running shoes for wet conditions?”

You might log:

• assistant: perplexity

• query_text: Best trail running shoes for wet conditions?

• brand_name: BrandX

• brand_mentioned: Y

• brand_rank: 2

• recommendation_flag: Y

• sentiment: positive

• pros_listed: strong grip; waterproof upper

• cons_listed: slightly heavier than others

• citations_present: Y

• citation_sources: brandx.com, rei.com

• competitors_named: BrandY, BrandZ

• purchase_intent_level: consideration

Common failure: Logging only mention counts (brand_mentioned) and ignoring rank, sentiment, and citations. That hides why you’re winning or losing recommendations.

3. Build prompt templates for each assistant

Different assistants have different strengths and answer styles. You’ll get better, more comparable data if you use consistent, assistant-specific prompts.

Your goal: Create repeatable prompt templates that can be reused weekly or monthly.

3.1 Universal structure for prompts

Each prompt should clearly specify:

• The task (e.g., list, compare, recommend)

• The context (e.g., region, budget, constraints)

• The output format (so it’s easy to log)

Generic template:

“You are helping a shopper in [REGION] looking for [CATEGORY]. They care about [KEY CRITERIA]. List the top options with pros and cons and clearly state which ones you recommend and why. Include citations to your sources.”

3.2 ChatGPT prompt templates

Category discovery (ChatGPT Search)

“A shopper in the United States is looking for the best [CATEGORY], with a budget of [BUDGET]. They care about [CRITERIA]. List your top 5 options with pros and cons, and clearly mark your primary recommendation. Include links to your sources.”

Branded decision (ChatGPT Search)

“Is [BRAND] a good choice for [USE CASE] in the United States? Describe pros, cons, and who it is best for. Please cite your sources.”

What to log: Whether ChatGPT names your brand, where it ranks in the list, how often it becomes the “primary recommendation,” and which sites it cites.

3.3 Perplexity prompt templates

Perplexity is naturally citation-heavy, which is ideal for tracking evidence.

Category discovery (Perplexity)

“For a shopper in [REGION], what are the best [CATEGORY]? Focus on [CRITERIA]. List at least 5 options with pros and cons, and highlight the top 1–2 you would recommend. Show your sources.”

Comparison (Perplexity)

“Compare [BRAND A] and [BRAND B] for [USE CASE] in [REGION]. Explain pros and cons, which is better for whom, and which one you would generally recommend. Please show the key sources you rely on.”

3.4 Claude prompt templates

Claude tends to provide nuanced reasoning. Make that explicit.

Branded decision (Claude)

“A shopper in [REGION] is considering [BRAND] for [USE CASE]. Evaluate whether this is a good option. Summarize pros, cons, and any important caveats. End with a single sentence stating whether you would recommend it and for which type of buyer. Use web sources and note them.”

3.5 Gemini prompt templates

Gemini’s strength is Google ecosystem grounding. Use that to see which sites drive your mentions.

Category discovery (Gemini)

“What are the best [CATEGORY] for a shopper in [REGION] who cares most about [CRITERIA]? List at least 5 products, with pros, cons, and a brief explanation of who each is for. Please show links to the main sources you’re using.”

Common failure: Changing prompts every time. That makes trend data impossible to interpret because the question keeps moving.

4. Run a repeatable measurement schedule

One-off audits are useful, but they don’t show whether your AI visibility is improving. You need a schedule.

Your goal: Move from ad-hoc checks to a repeatable measurement rhythm.

4.1 Choose your cadence

For most ecommerce brands:

• Weekly: Top categories and key markets

• Monthly: Long-tail categories, additional languages

Document your schedule:

• Week 1: US, English, top 3 categories

• Week 2: UK, English, same categories

• Week 3: DE, German, top 2 categories

• Week 4: Review and reporting

4.2 Assign ownership

Assign an owner for each assistant or region:

• Analyst A: ChatGPT + Perplexity, US and UK

• Analyst B: Claude + Gemini, EU

They are responsible for:

• Running the standardized prompts

• Capturing raw answers (copy/paste or export)

• Filling in the log table consistently

Common failure: Running all queries once, then never repeating them. You’ll never know if optimization efforts actually changed your AI share of voice.

5. Capture citations and sentiment consistently

Raw answers are helpful, but they only become analytics when you standardize how you interpret them.

Your goal: Turn every answer into structured data about citations, sentiment, pros/cons, and recommendation frequency.

5.1 How to capture citations

For each answer:

1. Check if citations exist

   • Set citations_present to Y or N.

2. Record the domains

   • Extract the top-level domains of cited pages.

   • Example: brandx.com, amazon.com, nytimes.com.

3. Classify citation type (optional but powerful):

   • citation_type_primary: owned (your site), retailer, review site, forum, media.

This helps you answer questions like:

• “Is Perplexity recommending us because of retailer pages or our own site?”

• “Which sites does Gemini lean on when it recommends competitors?”

5.2 How to label sentiment

Use a simple three-level sentiment scale:

• Positive: Clearly supportive language, strong or moderate recommendation

• Neutral: Balanced explanation, no clear recommendation either way

• Negative: Highlights issues, or explicitly advises against the brand

Example rules:

• If the assistant says, “BrandX is one of the best options…” → positive

• If it says, “BrandX is fine but not ideal for…” → neutral

• If it says, “BrandX is not recommended for…” → negative

5.3 Extract pros and cons

Capture up to 3 key pros and 3 key cons per answer:

• pros_listed:

  • waterproof; strong grip; good warranty

• cons_listed:

  • heavy; expensive

These become inputs for product and content teams:

• If every assistant lists “bad sizing” as a con, you know what reviews and Q&A to address.

Common failure: Treating citations and sentiment as “nice-to-have.” In reality, they are leading indicators of how models will rank and recommend you.

6. Track recommendation frequency and rank

Mention frequency is not the same as recommendation power. You need to know how often the assistants actually tell shoppers to choose you.

Your goal: Quantify how often each assistant recommends your brand and where you appear in their shortlists.

6.1 Define recommendation rules

For each answer, set recommendation_flag to Y if any of these conditions are true:

• The assistant explicitly says your brand/product is:

  • “one of the best options”

  • “a top recommendation”

  • “the best choice”

• Your brand is in the top N items (define N, often 3) in a “best of” list

Otherwise, set it to N.

6.2 Capture rank

When the assistant lists multiple options:

• Assign brand_rank based on order of mention.

• Example: If the list is BrandY, BrandX, BrandZ, then BrandX has brand_rank = 2.

This allows you to calculate:

• Average rank by assistant (e.g., ChatGPT vs Perplexity)

• Share of top-3 placements

• Change over time after optimizations

6.3 Simple metrics you can calculate

From your log, you can compute:

• AI share of voice:

  • brand_mentioned count / total answers

• Recommendation rate:

  • recommendation_flag count / total answers

• Average rank:

  • Mean of brand_rank for rows where brand_rank is not NULL

Common failure: Looking only at “did we show up?” and ignoring whether you’re actually selected in the buyer’s shortlist.

7. Turn logs into ecommerce and GEO/AEO actions

Tracking without action is just a dashboard. The point is to win more AI-driven recommendations and revenue.

Your goal: Use your AI assistant brand-mention logs to drive concrete ecommerce decisions.

7.1 Identify weak spots by assistant

Look for patterns like:

• ChatGPT mentions your brand but rarely recommends it

• Perplexity heavily cites competitor category pages, not yours

• Gemini recommends you only in one region or language

For each assistant:

• List top 3 queries where your rank or recommendation rate is poor.

• Inspect which domains are cited and which competitors are named.

7.2 Connect to catalog and content work

Use the insights to prioritize:

• Product page improvements:

  • Fill spec gaps that show up in pros/cons (e.g., weight, dimensions, materials)

  • Add comparison tables that match how assistants compare products

• Review and rating strategy:

  • If assistants cite retailer or marketplace reviews, target those platforms

• Content and GEO/AEO:

  • Create AI-optimized guides around the exact queries and decision criteria you see in answers

  • Ensure content is machine-readable (structured data, clean specs, consistent naming)

7.3 Operationalize with a platform

As your query set grows, manual logging becomes painful. This is where an AI visibility platform like Era can:

• Automate multi-model tracking across ChatGPT, Claude, Gemini, and Perplexity

• Capture citations, pros/cons, sentiment, and rank in a unified log

• Run ongoing GEO/AEO programs and content autopilot

• Provide CMO-ready reporting linking AI share of voice to ecommerce outcomes

Common failure: Treating this as an isolated SEO experiment instead of integrating it into your ecommerce and P&L-focused optimization roadmap.

FAQ: Troubleshooting Your AI Brand-Mention Tracking

1. What if different assistants give very different answers?

This is expected. Treat each assistant as its own observation layer. Instead of forcing a single combined score, track:

• Per-assistant AI share of voice

• Per-assistant recommendation rate

• Which data sources each model relies on

Then adjust your GEO/AEO work per assistant, starting with the ones most used by your customers.

2. How often should I update my prompts and queries?

Keep prompts stable for at least one quarter if you want trend data. Update them when:

• Your product catalog changes significantly

• You expand into new regions or languages

• You identify new, high-intent query patterns from search logs or customer research

When you do change prompts, note the change in your log so you don’t compare pre- and post-change data blindly.

3. What if assistants hallucinate or give factually wrong information?

Log it explicitly in the notes field. Then:

• Check whether your own site or key retailers provide clear, up-to-date data

• Ensure structured data and specs are consistent across sources

• Consider adding FAQ sections and clarifying content to reduce ambiguity

Hallucinations often signal weak or inconsistent machine-readable evidence.

4. How do I scale this beyond a few categories?

Start with a narrow set of high-revenue categories. Once your process works:

• Add new queries gradually

• Use automation (scripts, APIs, or a platform like Era) to:

  • Schedule prompts

  • Capture full answer texts

  • Populate your log schema automatically

5. How does this differ from traditional SEO rank tracking?

Traditional SEO tracks how a page ranks in search results. AI assistant brand-mention tracking measures:

• Whether your brand appears inside answers

• How it’s described (pros/cons, sentiment)

• Whether the assistant actually recommends you

It’s focused on the answer layer, which is increasingly where shoppers start, especially in conversational and agentic commerce.

By following these steps, you build an AI-native visibility system that helps you see—and improve—how Perplexity, ChatGPT, Claude, and Gemini talk about your brand. Over time, this becomes a core part of your ecommerce and GEO/AEO stack, not a one-off experiment.