← All guides · Published November 14, 2026 · Reviewed November 14, 2026 · By the Ingenious AI team · 6 min read
How Do You Prove What Your AI Block Kit Actually Blocked?
The short answer: with records, not intentions: a dated evidence pack built from the actual refusal events. AgentBlock's evidence tool parses server logs — including an offline parser that runs in your browser and never uploads the log — and the Team-tier export assembles the pack: total events, per-agent counts, paths and verdicts (BLOCK, RATE-CAP, SPOOF-SUSPECT, CANARY), stamped with the database version and generation date. That pack is what you hand a client, partner or insurer when 'we block AI training' needs to become a checkable claim.
Build my block kit — free · Test a policy in the simulator first
What is a block evidence pack?
a dated, exportable report of a kit's actual enforcement events — per-agent request and refusal counts with paths and verdict types — generated from parsed logs and stamped with the AgentBlock database version. The evidence page offers a local, offline log parser; the full pack export is a Team feature.
The claim that needs receipts
'We block AI crawlers' is currently closer to a vibe than a fact on most websites: a rule exists somewhere, nobody has counted its firings, and the count it reports includes phantom tokens and friendly fire. When the claim matters — a client contract names it, a data partner relies on it, an insurer prices it — the counterparty's next question is always evidentiary. Show the refusals: which agents, how many, on which paths, over what period, judged by which database version.
An evidence pack is that answer in one artifact. It converts enforcement from a property you assert into a record a third party can recompute from the same logs, which is the difference between a control and a claim about a control.
What goes into the pack
The pack is built by parsing real traffic records against the database: every classified event carries its agent, path, verdict and timestamp. Verdicts keep their types distinct — identity BLOCKs, RATE-CAP soft-blocks, SPOOF-SUSPECT claims, CANARY tripwire hits — because each type answers a different question a reviewer might ask, and flattening them into one 'blocked' number destroys exactly the information the pack exists to preserve.
The header states its own provenance: the database version the classification used and the generation date. A pack without a database version is unanchored — agent classifications change between versions, and a reviewer deserves to know which world the counts describe.
The offline parser, and why it matters
Server logs are sensitive: they carry visitor IPs, paths and timing — a map of your audience's behaviour. AgentBlock's evidence page therefore offers a parser that runs locally in the browser: the log is parsed on your machine and never uploaded. For many owners that property is the difference between evidence they can produce and evidence they cannot — the pack becomes possible without a data-handling conversation.
The Team-tier export then assembles the parsed picture into the dated report. Owners on lower tiers still have the plugin's block-log CSV export on WordPress and the Worker's structured events — the raw material of evidence is never gated; the pack is a convenience and a format, not the facts themselves.
Producing evidence on a rhythm
Evidence produced during a dispute is weaker than evidence produced on a rhythm — the same reason accounts are kept monthly rather than reconstructed at audit. Generate the pack monthly, file it beside the policy version and kit hash in force, and your crawler control acquires a history: refusal counts by month, verdict mix over time, the quarter the spoofing started, the month the fences began firing. Trends answer questions single packs cannot, including the only one that ultimately matters — is this control alive?
A flat-zero month in a rhythmic file is itself a finding: either the crawlers learned (possible, and visible in declining attempts at the edge of the window) or the artifact stopped enforcing (check with the verifier). Evidence that can embarrass you is the only kind worth filing.
Five actions to take today
- Parse locally when logs are sensitive. Use the evidence page's offline parser — the log never leaves your machine — for the classification pass.
- Keep verdict types separate. BLOCK, RATE-CAP, SPOOF-SUSPECT and CANARY answer different questions; never flatten them into one number.
- Stamp every pack. Database version and generation date belong in the header; unanchored counts cannot be rechecked later.
- File monthly, with the kit hash. Evidence plus the policy version in force is a history; a single pack is a snapshot that ages badly.
- Read a zero month as a finding. Verify the artifact with the kit verifier before concluding the crawlers simply gave up.
Worked example: the agency that stopped sending screenshots
Brightline Studio, a fictional web agency managing 14 client sites, used to answer 'is the AI blocking working?' with dashboard screenshots and reassurance. After moving its clients onto AgentBlock kits, it switched to monthly evidence packs: one per site, generated on the first Monday, filed with each site's kit hash.
The first cycle already paid for the habit. One client's pack showed zero events of any kind — the kit verifier then found the Worker route had been dropped in a hosting migration three weeks earlier. The client had been unenforced for most of a month, a fact the pack's silence revealed and the screenshot era would have missed entirely.
Across the other 13 sites, the packs showed 41,000 refusal events for the month with the verdict mix separated — including 900 SPOOF-SUSPECT claims on one retail client that opened, and then settled, a conversation about a price-scraping competitor.
Frequently asked questions
What is the difference between the evidence pack and a block log?
The log is the raw event stream — per-request records in the plugin's table or the Worker's structured events. The pack is the dated, aggregated report built from parsed logs: totals, per-agent counts, verdict mix, database version. Logs are evidence's raw material; the pack is the presentable, checkable form.
Can I build evidence without sending my logs anywhere?
Yes. The evidence page includes an offline parser that runs in your browser — local parsing never uploads your log. You classify on your own machine, then decide what, if anything, to export or share.
Which plan includes the evidence pack export?
The assembled pack export is a Team feature. The underlying facts are not gated: the WordPress plugin exports its block log as CSV, and the Worker's events are yours in your own logging. Team buys the format and aggregation, not the truth.
What should an evidence pack never claim?
That spoofed or unnamed scraping stopped, or that already-collected data was recalled. The pack reports classified events against a database version — named agents, verdicts, counts. Claims beyond the records are exactly what the pack format exists to prevent.
How long should packs be kept?
As long as the claims they support might be questioned — for client work, the life of the engagement plus your jurisdiction's contract horizon. They are small files; the expensive thing is the month you did not generate.
Related guides
- How to Block AI Scrapers and Bots from Your Website
- Which AI Crawlers Should You Allow?
- How Do You Block AI Bots on WordPress, Cloudflare, or Plain JS?
- Pre-deploy breakage simulator — test a policy before it ships
- Impact preview — your server log against a draft policy
- How Do Rate Limits Work for AI Crawlers?
- How Do Path Rules Protect Checkout, Login and Admin Pages from AI Agents?
- What Is a Canary Path and How Does It Catch Crawlers That Ignore Rules?
- How Can You Tell a Spoofed AI Crawler from a Real One?
- Which AI Agents Cannot Be Blocked by Name?
- Block evidence — parse a log, build an evidence pack
- The dated agent database — 25 entries, graded