How Gumshoe Works
API-based model access. Persona-driven testing. Statistical handling of non-determinism. Here is exactly how we produce AI visibility data you can trust.
Published Jun 2025 · Updated Mar 2026
Why Methodology Matters in a Non-Deterministic World
AI model responses are inherently non-deterministic. The same prompt, submitted to the same model, produces different answers at different times. Different brands get mentioned. Different sources get cited. Different narratives emerge.
This is not a bug. It is how large language models work. And it has a direct consequence for measurement: if your methodology does not account for non-determinism, your data is meaningless.
A single ChatGPT screenshot is not data. It is an anecdote. Building a GEO strategy on anecdotes is the fastest way to waste time and budget on the wrong priorities.
Everything in our methodology is designed to produce statistically reliable data from a fundamentally probabilistic system.
API Access: The Only Compliant Approach
Gumshoe accesses every AI model through official APIs. This is not a convenience choice. It is a compliance and data quality requirement.
Browser Scraping
- Violates most models' terms of service
- Data polluted by personalization, caching, and UI artifacts
- Breaks when providers change their UI
- Not auditable or reproducible
API Access (Gumshoe)
- Fully compliant with model provider terms
- Clean, structured data with no UI contamination
- Stable against provider UI changes
- Fully auditable and reproducible
Persona-Driven Testing: Context Changes Everything
A generic prompt like "What is the best CRM?" produces a generic answer. It tells you almost nothing about how AI models perceive your brand in the contexts that matter.
A persona-driven prompt like "I'm a VP of Sales at a 50-person B2B SaaS company. We need a CRM that integrates with Slack and handles complex deal stages. What should I evaluate?" produces a dramatically different — and far more useful — answer. Different brands appear. Different strengths get highlighted. Different sources get cited.
Persona-driven testing is not optional. Your brand does not have one visibility score. It has a different visibility profile for every buyer persona, on every model, at every point in time. Gumshoe measures all of them. Learn how to set up personas in the GEO 101 playbook.
How We Handle Non-Determinism
This is the hardest problem in AI visibility measurement. Here is how we solve it:
Breadth across prompts, personas, and models
Every report runs hundreds of conversations — unique prompts submitted once to each AI model. A brand that appears in 70 out of 100 conversations has 70% visibility. Spreading prompts across multiple topics, personas, and models is what captures natural variance in AI responses, rather than relying on any single output.
Statistical aggregation
Individual prompt results are aggregated across all prompts for a persona, across all personas, and across all models. The result is a visibility score that represents the probability of your brand appearing in AI responses for your target audience — not a cherry-picked example.
Timestamped snapshots for trend analysis
Every report is a point-in-time snapshot. By running reports on a regular schedule, you build a trend line that shows whether your visibility is improving, declining, or holding steady. This is how you measure the impact of your GEO efforts against the natural noise in AI responses.
Visibility Scoring: How the Numbers Work
Presence
Was your brand mentioned? The most basic signal: you either appear in the AI's response or you do not.
Prominence
Where were you mentioned? First recommendation or afterthought? Position and framing matter.
Sentiment
How were you described? Positive recommendation, neutral mention, or negative comparison?
AIO Page Audit: What We Measure
Beyond conversational AI, Gumshoe audits your presence in Google's AI Overview results. Five categories:
Content structure
How well your pages are structured for AI parsing — headings, lists, tables, schema markup.
Technical signals
Page speed, mobile readiness, crawlability, and structured data implementation.
Citation readiness
Whether your content is formatted in ways AI models prefer to cite. Learn more about citations.
Competitive positioning
How your pages compare to competitors for the same queries. See competitive benchmarking.
What Gumshoe Does Not Do
Transparency matters in a new category where vendors are making big claims. Here is what we do not do:
We do not scrape.
Every model interaction goes through official APIs. No browser automation, no headless browsers, no terms of service violations.
We do not guarantee placements.
No one can guarantee your brand will appear in AI responses. Anyone who claims otherwise is misleading you. We provide measurement and insights, not promises.
We do not game AI models.
No prompt injection, hidden text, or manipulation. These are unreliable, unethical, and counterproductive. We measure reality so you can improve it through legitimate means.
We measure and inform. You act.
Gumshoe is a measurement platform. We give you data and recommendations. Your team creates the content, builds the presence, and executes the strategy.
Methodology Summary
11
AI models tracked
API-only
Official model access
Persona-driven
Context-aware testing
Breadth
Many prompts × many models
Timestamped
Trend tracking
Transparent
No black boxes
See the methodology in action
Run a free report and see how Gumshoe measures your brand's AI visibility across 11 models.
Get Started FreeTrust the data. Own the narrative.
Start with a free AI visibility report. See exactly how AI models describe your brand.
Free to start · No credit card required