LLM Brand Visibility Scorecard
Measure how AI assistants perceive, mention, and differentiate your brand
- Difficulty
- Easy
- Time to result
- ~days to results
- Steps
- 6
- Confidence
- 96%
The LLM Brand Visibility Scorecard replaces keyword-ranking logic with a broader assessment of how AI assistants perceive and recommend a company. Start with structured brand inputs, then examine five signals: overall sentiment, share of voice, cited strengths, cited weaknesses, and recurring themes. Compare those signals with direct and adjacent competitors to identify whether the brand is prominent, differentiated, and positively represented. Because language models synthesize information from across the internet, their answers act as a rough proxy for accumulated public perception rather than a conventional search ranking. The output should drive focused action: select one or two weaknesses with the greatest reputational impact, improve the underlying experience or public evidence, and reassess over time and across several models.
Origin
Extracted from Marketing Against The Grain during a demonstration of HubSpot's AI Search Grader and its approach to measuring brand visibility in large language models.
Core principles
- 01Treat AI assistants as brand-perception engines, not traditional search engines
- 02Measure mentions and impressions alongside sentiment
- 03Compare brand visibility against direct and adjacent competitors
- 04Use recurring strengths and weaknesses to prioritize improvements
- 05Evaluate multiple language models because their results can differ
How to run it
- 1
Define the Brand Context
Provide the company name, location, product or service, and company type so the assessment targets the correct market context.
Pro tip Use the same description across repeated assessments so results remain comparable.
Watch out Ambiguous or overly broad inputs can produce irrelevant competitors and themes.
- 2
Measure Brand Sentiment
Review whether the model's discussion is positive, neutral, or negative and identify the specific experiences driving that sentiment.
Pro tip Separate emotional reputation signals from purely functional product commentary.
Watch out Do not treat one model's answer as a statistically representative customer survey.
- 3
Calculate Share of Voice
Compare how often the brand appears in relevant answers with how often competitors appear. Look for co-citation between the brand and important product or category terms.
Pro tip Include adjacent competitors when customers may consider them substitutes.
Watch out Low visibility may reflect a weak digital footprint rather than a poor product.
- 4
Extract Strengths, Weaknesses, and Themes
Group recurring observations into brand strengths, weaknesses, and differentiating themes. Verify whether those observations match real customer experiences and public evidence.
Pro tip Prioritize recurring findings that appear in several prompts or models.
Watch out Language models can reproduce outdated or inaccurate internet claims.
- 5
Choose Focused Improvements
Select one or two weaknesses whose improvement would most strengthen sentiment or differentiation. Address the underlying customer or product experience rather than merely changing promotional language.
Pro tip Customer support is a strong candidate when it repeatedly appears as a weakness.
Watch out Trying to improve every dimension simultaneously can dilute resources and accountability.
- 6
Compare Models and Reassess
Run the same assessment across OpenAI, Claude, Gemini, Perplexity, or other relevant assistants, then repeat after meaningful changes.
Pro tip Track results periodically to identify directional changes rather than reacting to a single snapshot.
Watch out Model updates can change results even when the brand itself has not changed.
In the wild
The hosts assess Coinbase and find low overall sentiment, with user friendliness, security, and regulatory compliance appearing as strengths. High fees, customer-support problems, and account restrictions emerge as weaknesses, while the share-of-voice comparison suggests Coinbase is underrepresented relative to several competitors.
→ The audit identifies customer support and fees as focused opportunities to improve brand sentiment.
A B2B software company runs the same brand prompts through three AI assistants each quarter. It records sentiment, competitor mentions, strengths, weaknesses, and recurring themes, then investigates findings that appear across at least two models.
→ The company gains a repeatable directional measure of AI-search visibility and a prioritized reputation backlog.
Common mistakes
Treating AI Results Like Keyword Rankings
AI assistants produce conversational synthesis rather than a stable list of ranked links. Measuring only rank misses sentiment, mentions, co-citation, and competitive context.
Overreacting to One Model
Different assistants may use different training data, retrieval systems, and response patterns. A reliable audit compares several models and repeated prompts.
Fixing Messaging Instead of Experience
Negative findings about fees, reliability, or support usually require operational improvements. Promotional copy alone will not create durable positive evidence across the internet.
Is it for you?
Best for
It is best for established or growing companies with enough online presence to appear in language-model training data.
Not ideal for
It is not ideal for new or offline businesses with almost no public digital footprint.
From the transcript
“what you really care about is like things like your share of voice”
“you can ask it things around what is the sentiment is my brand what are the good things the bad things how what's my share…”
“I always think it's easier to work on like one or two things than all of the things”
From the episode
Free AI Tool: See What The Internet Really Says About Your Brand