Imagine two people standing in a shoe store. One asks aloud, “Are Nikes good for marathon runners?” The other asks, “What’s the best running shoe?” Both want to buy, but they arrive with different expectations: one already knows the brand, the other is still deciding. The same split shows up in AI-powered answer engines — and it changes how brands should measure and act on what those systems say.
If your name appears only when people include it in the prompt, you’re visible to loyal customers but invisible to shoppers who haven’t heard of you yet. If you confuse those two audiences in your metrics, you’ll misunderstand where the real gaps are. Read on to learn how to separate these prompt types, run simple tracking, and draw useful conclusions from AI search behavior.
How to tell branded prompts from unbranded ones
A prompt that explicitly mentions a company, product, or a common misspelling is a branded prompt. Examples: “Are Nikes good for marathon runners?” or “Is HubSpot good for small business email marketing?” If the name is in the question, it’s branded.
An unbranded prompt contains no brand names. It focuses on category, problem, or use case: “What’s the best shoe for running a marathon?” or “How can small businesses improve email open rates?” Unbranded prompts are where discovery happens — where new buyers can learn about you for the first time.
Quick classification tips
– If a competitor’s name appears, treat that prompt as branded too — it’s not neutral.
– Product names count as brand mentions. A prompt that mentions a model, like “Air Force 1s,” is branded.
– If you can slot a prompt into one of these buckets in a few seconds, you’re ready for tracking.
Why the distinction matters for measuring AI visibility
Branded and unbranded prompts answer very different business questions. Mixing them into a single visibility score hides what’s really happening.
– Branded prompts reveal how AI describes your brand when someone already brings you up. They expose your positioning and whether details are accurate.
– Unbranded prompts show whether AI surfaces you when no one mentions your name — that’s true discovery and the tougher metric to move.
When you blend the two, you create a misleading “blended” visibility number. A high blended score can mask low discovery: many mentions might come from people who already know you, while strangers rarely see you in answers. That leads teams to allocate resources to the wrong fixes.
The blended-score problem — an example
Imagine you track 500 prompts in total: 100 branded and 400 unbranded. If branded answers mention you 90 times (90%) but unbranded answers mention you only 80 times (20%), your blended visibility is 170/500 = 34%. That 34% looks respectable, but it compresses two different realities: strong recognition among people who know you and weak discovery among newcomers.
A better approach: report the two rates separately and monitor the ratio between them.
Set up a prompt-tracking routine that actually works
AI-driven answers change over time and can vary between runs. A simple, repeatable tracking cadence reduces noise and surfaces trends.
Foundations to get right
– Build a deduplicated library: collapse prompts that ask the same thing in different words.
– Label each prompt: include status (active/paused/retired), buyer stage, branded vs. unbranded, geography/segment, and whether it names a competitor.
– Assign an ICP (ideal customer profile) and market: each prompt should map to a specific audience and region.
– Collect baseline data for at least a couple of weeks before making strategic calls.
Weekly tracking cadence
– Run prompts daily (or use a tool that does) to collect a distribution of answers.
– Review results weekly to identify gaps and avoid overreacting to single-day variation.
– For each group (branded, unbranded), calculate a visibility rate:
– Visibility Rate = (number of answers that mention your brand) / (total answers tracked for that group)
– Track empty or refused runs separately rather than treating them as zeros.
How to interpret the numbers and where to look next
Raw counts are tempting but often misleading. Always pair counts with clear denominators and focus on comparable rates.
Key signals to capture:
– Brand visibility: how often your brand is mentioned in answers.
– Prompt coverage: the share of tracked prompts where your brand appears at least once.
– Share of voice: your brand’s mentions vs. total brand mentions (you + competitors) across the same prompts.
– Citations: the sources the engine references when producing an answer; those may or may not mention you.
How to read the branded-to-unbranded ratio
– Ratio = (branded visibility rate) / (unbranded visibility rate)
A high ratio means AI mentions you mostly when people already know your name. A declining ratio implies improved discovery. If asked for one headline number, lead with the unbranded rate — it’s the harder-to-move measure and most predictive of new-customer discovery.
A practical reporting checklist
When you prepare your weekly analysis, include:
– Date range, answer engines used, and number of answers behind each rate.
– The denominator for every percentage (answers or prompts).
– Branded and unbranded rates reported separately.
– The branded-to-unbranded ratio and its trend over several weeks.
– Any prompt-library changes since the last report.
– Competitor share of voice for unbranded prompts.
– Top-cited domains and which are third-party.
– Note limitations: variability of answers and any refused/empty runs.
– One specific decision or action tied to each major finding.
Also: open a sample of captured answers (5–10) and inspect how you’re described, which sites are cited, and whether the description is accurate and favorable. Keep a short “what we changed” log next to your weekly numbers so you can correlate actions with movement in visibility.
How to implement a small, practical tracking system
You don’t need fancy software to start — a spreadsheet can work for a small set of prompts — but scale will make manual tracking painful.
Steps to begin:
– Choose a manageable set: pick a handful of branded and a handful of unbranded prompts per buyer stage.
– Run them across 1–3 answer engines (for example ChatGPT, Gemini, Perplexity).
– Collect results daily for at least two weeks, then review weekly.
– Label and store the exact answers so you can audit any changes.
– Calculate and report the two visibility rates and the branded-to-unbranded ratio.
When spreadsheets start to fail
If you’re tracking dozens of prompts or receiving hundreds of engine responses, manual handling becomes slow and error-prone. At that scale look for tooling that can:
– Run prompts automatically across multiple engines.
– Store and display exact answers alongside the metrics.
– Break citations down by domain and content type.
– Provide filters by buyer stage, product, and geography.
How product-level AEO tools can help (what they typically do)
Tools built for answer-engine optimization generally align with this workflow:
– Suggest prompt ideas based on your products and customer profiles.
– Help organize prompts into groups and apply labels (stage, region, branded/unbranded).
– Run prompts on a schedule across supported answer engines.
– Show the exact responses each engine returned for each prompt.
– Calculate Brand Visibility, prompt coverage, sentiment, and trend over time.
– Surface competitor mentions and share of voice for the same prompts.
– Break citations out by top domains and content types and translate patterns into prioritized actions.
If you adopt a tool, keep a naming convention for branded vs. unbranded prompts so you can pull separate rates. The tool should show both the metric and the answer behind it — that makes verification quick before you share numbers.
Common questions teams ask when they start tracking
How many prompts should I create per buyer-journey stage?
Start small. A handful per stage will reveal patterns without being overwhelming. If tool limits constrain you, distribute prompts evenly across stages and split each stage between branded and unbranded.
How often should the prompt library be refreshed?
Review outputs weekly, QA the library monthly, and refresh prompts quarterly or whenever your products, audiences, or competitors change.
Should product names appear in unbranded prompts?
No. Product names are brand indicators; include them in the branded set. Keep unbranded prompts strictly about category, problem, or use case.
How should I handle different geographies or segments?
Create separate prompts and labels for each market or segment. Don’t mix them: answers and citation patterns often vary by region.
What if an engine refuses to answer or returns empty runs?
Log refusals as gaps, not zeros. If an engine consistently refuses a prompt, consider rewriting it more neutrally or retiring it. Document refusals and state whether you excluded them from denominators in your reports.
Practical first moves you can take this week
– Pick 6–10 prompts today: split them between branded and unbranded and across top-of-funnel and consideration stages.
– Run those prompts across one or two answer engines and save the responses.
– Calculate the two visibility rates and the branded-to-unbranded ratio for that first batch.
– Read 5 captured answers to check descriptions and citations.
– Note one change you’ll try (content refresh, FAQ addition, or targeted third-party outreach) and who will own it.
Similar Posts
- AI Search Performance Benchmark: How to Compare Your System Against Industry Standards
- AEO Prompt Tracking: Optimal Frequency to Run Prompts for Reliable Results
- AI Prompts You Must Track: A Practical Guide
- AEO & SEO: Protect Search Visibility as Buyers Move Beyond Traditional Search
- Master AI Search: Essential Tips to Monitor Your Brand’s Visibility
Mason Brooks is a blockchain business strategist with six years of experience launching crypto startups in the United States. He breaks down innovative business models and best practices to turn your ideas into reliable revenue streams. His pragmatic approach provides you with actionable roadmaps and inspiring case studies.
