Editorial rules
Benchmarks and reviews should start from saved observations: task sets, URLs, responses, latencies, errors, receipts, screenshots, or reproducible notes.
Search APIs, SERP wrappers, extraction APIs, CLIs, MCP servers, and SaaS products should not be collapsed into one misleading leaderboard.
A tool can be best for official-docs retrieval, weak for fresh news, and unsuitable for write actions. Recommendations should say where the tool fits.
Benchmarks that depend on changing web results, prices, models, or APIs should be labeled with the run date and enough evidence to understand the snapshot.
Good tools should be included even if they have no affiliate programme. A paid relationship cannot override the evidence.
If a result, price, API behavior, or recommendation becomes wrong, the page should be corrected or marked stale instead of quietly preserving a convenient claim.
Affiliate and referral links
Some pages may contain affiliate or referral links. If a reader signs up or buys through those links, AgentFirstTools may receive a commission, credit, discount, or other benefit at no extra cost to the reader.
- Disclosure should appear near the relevant recommendation, not only hidden on this page.
- Affiliate links should not be cloaked as evidence. The reason for recommending a tool should be visible in the review, benchmark, or comparison.
- Non-affiliate tools remain eligible. If the best answer does not pay, the page should still say so.
- Commission size should not determine order, scores, badges, or conclusions.
Sponsorships and vendor relationships
AgentFirstTools may accept sponsorships, vendor-provided credits, demo access, research funding, or paid vendor audits where that helps produce useful evidence. Those relationships must be labeled clearly when they could affect reader trust.
Sponsored placement is separate from a recommendation. A sponsor may be visible on a page, but cannot buy a top ranking, “best overall” verdict, or removal of negative findings.
Paid audits and buyer reports
AgentFirstTools may sell private tool-selection audits, benchmark scoping, buyer reports, implementation playbooks, templates, or workflow setup help. Those services should use the same agent-first criteria as the public site: inspectability, scriptability, bounded action, verification, recovery, and composability.
Private work should not silently distort public rankings. If private work produces a public finding, sensitive details should be removed and any relevant commercial relationship should be disclosed.
What counts as good evidence?
- Task-shaped benchmark inputs and dated cohorts.
- Saved response snippets, URLs, status codes, result counts, latency, and error behavior.
- Concrete workflow traces that show what an agent could inspect, call, verify, retry, or recover from.
- Clear limitations where sample size, changing web results, unavailable credentials, or manual judgment reduce confidence.
- Reusable artifacts such as CSVs, JSON evidence bundles, scorecards, checklists, command examples, schemas, or receipts.
Corrections and review
If a page contains an error or a vendor believes a benchmark misrepresents its product, AgentFirstTools should review the underlying evidence and update the page where appropriate. Corrections should improve accuracy without hiding the original limitation of a dated run.
For commercial or correction inquiries, use the relevant inquiry form on the audit or benchmark pages rather than sending secrets, credentials, customer data, or confidential logs.