Prompt
I'm building a SaaS that needs to show if our brand appears in ChatGPT and Perplexity for category queries. What would you track?
Latest observation
If I were building this, I’d track visibility at the query level, not just “are we mentioned somewhere.” For ChatGPT and Perplexity, the useful unit is:
Category query → answer → brand presence/placement/context
Here’s the framework I’d use.
1) Core visibility metrics
A. Share of Voice in AI answers
For a set of category queries, measure:
- Mention rate: % of answers where your brand appears
- Rank/position: where your brand appears in the answer list
- First mention rate: % where you’re the first brand mentioned
- Top-3 / Top-5 inclusion rate
- Average position across responses
This is the closest analog to SEO rank tracking.
B. Citation / source visibility
Especially for Perplexity, but also useful for ChatGPT when citations are available:
- Citation inclusion rate: how often your site or owned content is cited
- Citation position: whether your citation supports the main recommendation or is buried
- Citation type: your site, third-party review, listicle, forum, social proof, docs, etc.
- Competitor citation rate
- Authority of cited sources
This tells you whether the model is “learning” about you from your own content or from others.
C. Recommendation quality
A mere mention isn’t enough. Track:
- Recommended vs. merely mentioned
- Neutral vs. favorable vs. unfavorable sentiment
- Use-case fit: is the brand recommended for the right segment?
- Feature accuracy: are the model’s claims about your product correct?
- Category relevance: is your brand mentioned in the right category, or being miscategorized?
2) Query coverage metrics
You need a query set that reflects how buyers actually ask.
A. Query types
Track by intent buckets:
- Generic category: “best project management software”
- Use-case specific: “best project management software for agencies”
- Comparison: “Brand A vs Brand B”
- Problem/solution: “software for managing client approvals”
- Budget-led: “cheap CRM for startups”
- Enterprise/security/compliance
- Migration/alternative queries: “alternatives to X”
B. Coverage
For each bucket:
- % of tracked queries where you appear
- % where competitors appear and you don’t
- Query segments with highest/lowest visibility
- Long-tail coverage
- Brand mention frequency by intent
This shows where you’re strong or weak.
3) Competitive metrics
A. Relative share
Track against a competitor set:
- Mention share vs competitors
- Average rank vs competitors
- Co-mention frequency: how often you appear with specific competitors
- Winner rate: how often the model chooses your brand as the top recommendation
B. Competitive displacement
Important question: when competitors appear, what are they being recommended for?
- Feature-based displacement: competitor wins on pricing, integrations, etc.
- Segment-based displacement: competitor wins for SMB vs enterprise
- Source-based displacement: competitor is cited from stronger third-party coverage
4) Answer mechanics metrics
These help you understand how the model is constructing responses.
A. Answer format
- Listicle / ranked list / narrative / comparison table / direct recommendation
- Does the answer include explicit criteria?
- Does it ask clarifying questions instead of answering?
B. Determinism / variability
Because LLM outputs vary:
- Run each query multiple times
- Track response stability
- Confidence of mention: how consistently your brand appears across runs
This is critical for trustworthy reporting.
C. Prompt sensitivity
Track performance across:
- Different phrasings of the same query
- Different locales
- Different devices/sessions if relevant
- Different recency conditions
5) Brand perception metrics
You want to know not just if you’re present, but how the AI frames you.
A. Attribute association
Track whether the model associates your brand with the attributes you want:
- Best for teams / enterprise / startups
- Cheapest / easiest / most secure / most accurate
- Integrations / automation / analytics / support
B. Misrepresentation detection
- Wrong pricing
- Outdated features
- Wrong target customer
- Confusing you with another brand
- Claiming unsupported capabilities
C. Category ownership
Do buyers see you as:
- A leading brand in the category?
- A niche player?
- An alternative?
- A tool for a subsegment only?
6) Source ecosystem metrics
For Perplexity especially, and indirectly for ChatGPT answers grounded in web content:
A. Web presence signals
Track whether your brand appears in:
- Review sites
- Comparison articles
- Reddit / forums
- YouTube / podcasts (if surfaced)
- Your own docs/blog/landing pages
- G2/Capterra/alternatives pages
B. Source quality balance
- Owned vs earned vs third-party citations
- Freshness of sources
- Source diversity
- Presence in high-authority sources
This helps explain why you’re being surfaced or not.
7) Practical KPI dashboard I’d build
If this were my product, the core dashboard would show:
-
Visibility score
- Weighted mention rate across target queries
-
Share of voice
- Your brand vs competitors across all tracked queries
-
Top query gaps
- Queries where competitors are present and you’re absent
-
Top opportunity queries
- High-volume/high-intent queries where you’re close to appearing but not yet included
-
Citation health
- Which sources drive inclusion, and whether your own content is cited
-
Position trends
- Month-over-month movement in rank/mention rate
-
Brand sentiment / accuracy
- Positive, neutral, negative, and misinformation flags
-
Query intent performance
- By segment and use case
8) How I’d structure the data model
For each query-response pair, store:
- Query text
- Intent bucket
- Platform: ChatGPT / Perplexity
- Locale / language
- Timestamp
- Prompt variant
- Response text
- Brand mentions
- Brand position
- Competitors mentioned
- Citation URLs
- Citation domains
- Sentiment
- Accuracy flags
- Response format
- Stability score across repeated runs
This gives you enough structure to support both dashboards and alerts.
9) Important caveat: “presence” is not enough
A brand being mentioned in an answer can be:
- a recommendation,
- a warning,
- a comparison loser,
- or just incidental context.
So I’d always pair visibility with:
- role in the answer,
- sentiment,
- accuracy,
- and whether the answer matches commercial intent.
10) If you want the shortest possible set of metrics
If you only tracked 6 things, I’d do:
- Mention rate
- Top-3 inclusion rate
- Average position
- Citation rate
- Competitor share of voice
- Answer sentiment/accuracy
If you want, I can also turn this into:
- a schema for your database,
- a dashboard layout,
- or a scoring formula for an “AI visibility score.”