What counts as a trust receipt?
By receipt, I mean a concrete piece of evidence that lowers uncertainty before purchase. A public pricing page counts. So does a sample output, a changelog, an export button, a security page, a refund policy, a model card, or a visible limitation.
For AI tools, receipts matter because the demo is often easier than the durable product. A model can generate a convincing mockup in seconds; building reliable workflows, safe permissions, good billing, and clear support takes much longer.
Six receipts I ask for
| Receipt | Buyer question | Good evidence |
|---|---|---|
| Pricing receipt | What will this cost after the trial? | Plan limits, usage units, overage rules, cancellation terms, and whether API/model costs are included. |
| Product receipt | Does the product do a real job? | Annotated screenshots, sample projects, before/after examples, and a workflow that starts and ends outside the AI box. |
| Output receipt | Can I judge the result without becoming the product's QA team? | Confidence labels, sources, diffs, citations, logs, structured exports, or review queues. |
| Data receipt | What happens to the data I provide? | Retention policy, training policy, region notes, deletion controls, and a clear list of subprocessors or model providers. |
| Control receipt | Can I stop, undo, approve, or audit actions? | Permission scopes, approval steps, change history, admin controls, and activity logs. |
| Exit receipt | What happens if I leave? | Export formats, data deletion steps, billing cancellation, and whether generated work remains accessible. |
The receipts change by category
A chat assistant, an image generator, a coding agent, and a business automation agent do not need the same evidence. The risk surface changes with what the tool can touch.
| Category | Most important receipt | Reason |
|---|---|---|
| Chat assistant | Data and output receipts | You need to know what is retained and how to verify answers. |
| Coding assistant | Control and output receipts | The main risk is unreviewed code, wrong diffs, or hidden security changes. |
| Image generator | Pricing and rights receipts | You need to know cost per output, reuse rights, and export quality. |
| Agent workflow tool | Permission and audit receipts | The tool may read data, write records, contact people, deploy changes, or spend money. |
| API provider | Pricing and reliability receipts | Small per-call differences become large bills at scale; downtime can block your product. |
Red flags that make me pause
- No visible pricing: acceptable for enterprise sales, risky for individual or small-team tools.
- No sample outputs: especially bad for tools selling summaries, code, research, or analysis.
- No cancellation explanation: a cheap trial can become expensive if exit is hidden.
- No data policy near the product: a legal page buried in the footer is not enough for sensitive workflows.
- Only benchmark claims: benchmarks can be useful, but they rarely describe your workflow.
- Unlimited language with limited details: "autonomous", "agentic", and "enterprise-grade" should come with controls.
A due diligence loop for a small team
You do not need a procurement department to do useful diligence. For a personal or small-team tool, I would keep the loop simple:
- Read the pricing page first. If the pricing model is unclear, do not start a serious trial yet.
- Run one non-sensitive task. Use realistic input, but avoid private code, customer data, or confidential files.
- Evaluate the output as a cost. Count review time. If every output needs heavy repair, the tool is not saving work.
- Find the delete/export/cancel path. Do this before you depend on the tool.
- Write a one-paragraph internal note. State when to use it, when not to use it, and what data is allowed.
The receipt score I would use
I would give one point for each receipt I can find before paying: pricing, product, output, data, control, exit. A consumer tool with 4 or more is usually ready for a cautious trial. A team tool that can touch code, money, customers, or production should have 5 or 6 before serious use.