AI Visibility Methodology for SaaS | ai-visibility.best

What This Methodology Is For

ai-visibility.best measures observable AI answer patterns for SaaS buying prompts. The goal is not to pretend that AI search has a perfect ranking API. It does not. There is no public Search Console for these six AI platforms that tells a SaaS founder every impression, prompt volume, or personalized answer a buyer saw.

So we measure the evidence a team can actually inspect: the prompt, the platform, the timestamp, the raw answer, the brands mentioned, the sources cited or implied, and the fix that is most likely to make the product easier for AI systems to understand.

This is a practical methodology for founder-led SaaS teams that need to answer three questions:

• Do AI assistants mention or recommend our product for the buyer prompts that matter?
• Which competitors appear instead, and what sources seem to support them?
• After we change a page, publish a comparison, improve docs, or earn a source, did the same prompt set move?

What We Measure

An ai-visibility.best scan tracks whether a SaaS product is present in AI answers for buyer-intent prompts. We classify each answer into three plain-language states:

We also record competitor mentions, answer prominence, cited sources, source gaps, description accuracy, and suggested fixes.

Prompt Sampling

Each project starts with the product name, website, category, positioning, known competitors, and visible buyer use cases. From that, ai-visibility.best builds a prompt set that reflects how a buyer researches software, not how a vendor talks about itself.

The prompt mix includes:

• Best-in-category prompts, such as best project management tools for small teams.
• Use-case prompts, such as best analytics software for privacy-focused SaaS.
• Alternative prompts, such as [competitor] alternatives.
• Comparison prompts, such as compare [product] vs [competitor].
• Scenario prompts, such as which [category] tool should I use for [job].

The same prompt set is saved and reused over time. That matters because one-off manual checks are easy to misread. A founder needs to know whether the same buyer questions are improving after specific changes ship.

Platforms and Access

Paid scans query six pinned models through OpenRouter: ChatGPT, Claude, Gemini, Perplexity, Grok, and DeepSeek. The free preview queries ChatGPT only. Every paid run records the exact model, raw answer, timestamp, and citations returned for that platform.

Each scan runs six saved buyer prompts across all six platforms, producing an expected matrix of 36 observations. Evidence records label retrieval provenance as native, built_in, or exa. These labels describe the measured API route, not consumer-app browsing behavior.

We treat platform results as comparable evidence, not identical experiences. Depending on the platform, scans may use available APIs, answer endpoints, or controlled query methods. This creates a repeatable proxy for the answer pattern, but it is not a pixel-perfect copy of every signed-in, personalized app session on a buyer's device.

That limitation is important. We would rather show repeatable evidence with clear caveats than claim false precision.

Platform and crawler behavior is checked against primary documentation where available, including OpenAI crawler documentation, Perplexity crawler documentation, and Google guidance for AI features in Search.

Run-to-Run Noise

AI answers are nondeterministic. The same prompt can return different wording, order, cited sources, or brand lists across runs. A single snapshot can make a product look safe one day and invisible the next.

ai-visibility.best handles this in four ways:

1. Every run stores the raw answer and timestamp.
2. Repeated scans use the same prompt set so teams can compare like with like.
3. Movement is interpreted as a pattern, not a guarantee from one run.
4. Reports separate per-platform evidence instead of hiding disagreement behind one blended score.

This is why the dashboard is built around prompt evidence, source gaps, and retesting, not just a vanity score.

How the Visibility Score Works

The score is a summary of the evidence, not the evidence itself. Every score should be opened and checked against the prompts and raw answers behind it.

Only grounded provider_api observations with successful normalization are scored. Provider errors, ambiguous matches, mock rows, and ungrounded answers never become brand absence and never enter the score.

Within each platform, recommendations receive the most credit, with a small rank discount when the product is buried; mentions receive half credit; competitor wins and not-found results receive zero. Incorrect or outdated descriptions are discounted. Buyer-intent prompts carry more weight than lower-intent discovery prompts.

The six platform scores are then averaged equally. A platform with more completed rows cannot outweigh another platform.

The numeric score remains hidden until all of these publication conditions pass:

• At least 33 of 36 grounded observations are valid.
• Every platform has at least five of six valid observations.
• All three critical buyer/comparison prompts are valid on every platform.

At 33-35 valid observations, the report is labeled measured. At 36/36 it is final. Below the gate, the dashboard shows coverage and provider errors without publishing a number.

The score is useful for trend direction. It is not an official market-share number, exact rank, or search-volume estimate.

Source Gap Analysis

When a competitor wins a prompt, the useful question is not just "who appeared?" It is "what evidence did AI have that made that competitor easier to recommend?"

ai-visibility.best looks for source gaps across:

• Comparison and alternatives pages.
• Review sites such as G2 and Capterra.
• Reddit, community threads, and founder discussions.
• Clear homepage positioning and category language.
• FAQ, documentation, integration, and use-case pages.
• Structured data and machine-readable summaries.

The output is a fix backlog: pages to improve, comparisons to create, sources to earn, copy to clarify, and prompts to retest.

What Every Report Should Let You Inspect

A trustworthy AI visibility report should not ask you to trust a black box. For every important result, ai-visibility.best is designed to expose:

• The exact buyer prompt.
• The platform tested.
• The timestamp.
• The raw answer excerpt.
• The detected status: Found, Knows You, or Not Found.
• Competitors mentioned or recommended.
• Sources cited or source gaps found.
• The suggested fix and the reason for that fix.

If a classification looks wrong, the evidence should be visible enough for a founder or marketer to challenge it.

What We Do Not Claim

ai-visibility.best is intentionally conservative about claims:

• We do not guarantee ChatGPT, Claude, Gemini, Perplexity, Grok, DeepSeek, or any AI assistant will recommend your SaaS.
• We do not provide exact prompt volume or total AI demand.
• We do not claim Search Console-level accuracy.
• We do not claim our proxy matches every signed-in, personalized session.
• We do not replace SEO analytics, product analytics, or sales attribution.

The value is not certainty. The value is repeatable evidence, competitor displacement, source gaps, and a short list of fixes worth shipping.

How We Handle Your Data

Scans use the public website, product name, category, competitors, and AI answers generated from the prompt set. ai-visibility.best does not need private revenue, customer, CRM, or analytics data to run an AI visibility check.

Reports are built to help SaaS teams decide what to fix next. We do not sell scan data, and teams should be able to delete their projects and scan history from their account.

See the Method Applied

The methodology is easier to judge when applied to your own product. Run a free AI visibility check, open the evidence behind each result, and decide whether the prompt set, raw answers, source gaps, and fixes are useful.

Run Free AI Visibility Check

Read the SaaS AI Visibility Benchmark

Primary References

These references define the technical standards used in page-readability and structured-data checks. They do not validate or endorse ai-visibility.best scores.

• Google Search Central: AI features and your website
• Google Search Central: Dataset structured data
• OpenAI: crawler and user-agent documentation
• Perplexity: crawler documentation
• Schema.org Dataset specification