AI buyer questions and answer evidence

What this evidence directory records

Slate records the buyer questions used to evaluate how B2B software appears in AI-assisted discovery. Each published category page connects a real question to the public answer returned by an AI assistant, the products named in that answer, and the external source domains cited alongside it. This makes the evidence inspectable: buyers can see more than a leaderboard score, while software teams can examine the language and sources shaping recommendations. Prompt evidence supports the category rankings, but it is also useful on its own for understanding how different assistants frame a market and which products enter a buyer’s initial consideration set.

How buyer questions are selected

Questions belong to versioned, category-specific prompt packs. They are written around practical discovery and evaluation needs, such as identifying suitable products, comparing options, or finding software for a particular workflow. A Helpdesk question is evaluated against the active Helpdesk roster; it is not mixed with products from an unrelated category. Versioning keeps the prompt set aligned with the roster and the promoted ranking snapshot. When a category or question set changes, the resulting evidence is tied to that version rather than silently presented as if the underlying inputs stayed the same.

The directory publishes category questions, not personalized recommendations. The prompts are designed to observe assistant behavior consistently across a defined software market.

Which AI assistants are covered

The current public evidence model covers ChatGPT, Claude, Gemini, and Perplexity. Slate runs eligible category questions across the configured assistants and stores the completed public response for each platform when it is available. Answer cards remain platform-specific because assistants can name different products, order them differently, or cite different sources for the same buyer question. Showing those differences is more useful than merging every response into one artificial consensus. A missing answer is displayed as pending; the page does not invent a response or transfer evidence from another assistant.

What must pass before a category is published

Prompt pages are gated separately from ordinary category pages. A category can have a useful product roster or leaderboard while its prompt page remains unavailable to search engines. Slate publishes a prompt page only when the category is active, its category page is published, the supporting snapshot is promoted, and the evidence clears the fixed answer-and-source threshold. Runs identified as Slate self-citations are excluded from this public evidence decision.

• The active prompt pack must contain at least five buyer questions.
• At least half of those questions must have a completed public answer.
• At least three questions must contain eligible external source evidence.
• The qualifying evidence must include at least three distinct external source domains.

How to read an answer card

Start with the question, then compare the assistant cards rather than treating one response as definitive. A product chip means that the product was explicitly matched to the published category roster for that response. Its displayed position describes where it appeared in that answer; it is not a market-share estimate, customer rating, or permanent category rank. Source chips show the public domains associated with the stored response. They help explain which publishers or official sites informed the answer, but a citation does not mean Slate endorses every claim on that domain.

Use the linked product profile for product details and the category leaderboard for aggregated visibility metrics across eligible runs. The prompt page is the evidence layer connecting those higher-level results back to individual buyer questions.

Freshness and limitations

AI answers can change when models, retrieval systems, public sources, prompts, or product rosters change. Each published category therefore reflects a specific promoted snapshot and data period rather than a timeless recommendation. Publication dates indicate when evidence entered the public inventory; they do not guarantee that every cited page remains unchanged afterward. Slate excludes incomplete and self-cited evidence from the publication threshold, but the visible results are still a directional record of assistant behavior, not proof of product quality or buyer preference.

A product may be absent because it was not mentioned in the eligible answers, because it was outside the approved roster, or because the public evidence was incomplete. Review the connected category, product, and source pages before drawing a purchasing or competitive conclusion.