← All guides · Published November 14, 2026 · Reviewed November 14, 2026 · By the Ingenious AI team · 7 min read

How Should an Agency Report AI Crawler Control to Clients?

The short answer: report artifacts, not assurances. The package has five pieces, all generated by the work itself: the client's policy with its per-agent decisions and recorded exceptions; the deployed kit's ID and hash, verified; a monthly evidence pack — per-agent counts with BLOCK, RATE-CAP, SPOOF-SUSPECT and CANARY verdicts kept separate, stamped with the database version; a note whenever a database diff changes anything for their site; and a quarterly review where the numbers, not the agency's confidence, carry the conversation.

Build my block kit — free · Test a policy in the simulator first

What is agency crawler reporting?

the recurring evidence package an agency delivers for managed AgentBlock policies: the graded policy record, verified kit fingerprints, monthly evidence packs with separated verdict types, database-change notes, and a dated exceptions list — designed so a client can check every claim the report makes.

Why assurances stopped working

Clients have been sold 'AI protection' by vendors whose deliverable was a screenshot. They arrive at an honest agency pre-disappointed, and they ask the question this guide answers in the form this paragraph is written in: how would I know? A reporting package is the answer's durable form — a set of artifacts each of which the client, or the client's next agency, or the client's auditor, can independently check. Checkability is the product; the monthly PDF is just its envelope.

The package also protects the agency. When a client's citation traffic dips for reasons that live in an answer engine's update, the fetch counts in the evidence file clear the policy of suspicion in one reading — the same instruments, working for the defence, that everywhere else in this series work for the diagnosis.

The five artifacts, in the client's language

One, the policy: who is refused, who is allowed, what is fenced — written per agent with reasons a business owner can follow, exceptions named. Two, the fingerprint: kit ID and hash, with the verification result and date, proving the deployed artifact is the policy on paper. Three, the monthly counts: refusals and verdicts per agent from the client's own traffic, with the verdict types separated because a client who learns to distinguish a SPOOF-SUSPECT from a BLOCK has understood their estate better than most vendors do. Four, change notes: every database version that touched their posture, what the diff contained, what the agency decided. Five, the exceptions list: every deliberate deviation from the house policy, with its reason and review date.

Delivered monthly as a rhythm and quarterly as a conversation, the package replaces the two failure genres of agency reporting — the data dump nobody reads and the reassurance nobody can check — with a file that accrues into the client's diligence record.

Numbers a client can actually use

Choose the figures for the decisions they inform. Training refusals quantify the inventory protection. Retrieval fetches beside answer-referred sessions price the citation channel — the pair belongs on the same page, always, because together they are the policy's profit and loss. Path-rule firings answer the security question in the form payment providers ask it. SPOOF-SUSPECT counts open the competitor-scraping conversation at the right temperature: evidence first, escalation proportionate. CANARY hits are reported as what they are — detections, not accusations.

What never appears: an ungraded 'agents blocked' total, a percentage with no denominator, or a claim about data already collected. Clients forgive a small honest number instantly; they remember a large decorated one at renewal.

The quarterly conversation

The quarterly review runs on the file the monthly packs built: trends in each verdict class, the substitution question for retrieval with the client's freshest referral data, caps that bound and what they reported, the database versions absorbed and the decisions they required, exceptions re-justified or retired. Decisions taken in the meeting are flips in a builder and a regeneration — small, dated, verified — and the next quarter's pack opens by confirming them in traffic.

That closing of the loop is the retention mechanism agencies actually have. A client who watches their own numbers change because of decisions they remember making does not churn over a dashboard's colour scheme.

Five actions to take today

  1. Report artifacts, not assurances. Policy, fingerprint, evidence pack, change notes, exceptions — five checkable pieces, every cycle.
  2. Verify before you report. The kit hash check precedes the monthly pack; never report a posture you have not confirmed is deployed.
  3. Keep verdict types separate. BLOCK, RATE-CAP, SPOOF-SUSPECT and CANARY answer different client questions; flattening them destroys the report's value.
  4. Pair retrieval fetches with referrals. The policy's profit-and-loss on one page: what the citation channel costs in fetches and pays in sessions.
  5. Log decisions in the meeting, ship them the same week. Flips, regeneration, verification, and the next pack confirms — the loop is the service.

Worked example: the renewal meeting that ran on rails

Brightline Studio's largest client, a fictional 3-location retailer, came to its annual renewal with a new operations manager whose first question was the standard one: 'What exactly have we been paying for on the AI side?' The studio opened the file instead of a slide deck: twelve monthly evidence packs, 212,000 refusal events across the year with verdict types separated, four database versions absorbed with their diffs and decisions minuted, two exceptions — a partner fetcher allowlisted in March, a fence widened in July — each with its reason and the traffic that justified it.

The operations manager spent the meeting on the two pages that mattered to her: retrieval fetches against answer-referred revenue (the channel was paying for the service several times over) and the SPOOF-SUSPECT series that had identified a competitor's scraping in May, with the fence change that ended it.

Renewal was signed in the meeting, at a higher tier. The studio's account lead attributes it to a sentence the client said on the way out: 'You're the first vendor whose file I could hand to our auditors unchanged.'

Frequently asked questions

What should a client report never contain?

Ungraded totals ('31 AI bots blocked!'), claims about already-collected data being recalled, promises that spoofed scraping has stopped, and screenshots in place of records. Each is a claim the file cannot cash, and a sophisticated client reads their presence as the report's true content.

How technical should the client-facing version be?

Name the agents, keep the classes, translate the stakes. Clients follow GPTBot-versus-ChatGPT-User perfectly well when the citation consequence is attached; what loses them is verdict vocabulary without the 'so what'. The evidence pack's numbers stay exact — it is the framing that gets translated, never the counts.

Can the client verify the kit themselves?

Yes, and the offer should be made: the kit ID and hash are in their policy record, the verifier is a public page, and the WordPress clients hold their own block log in their own database. A client who has personally verified once renews differently from one who has only been assured monthly.

What cadence do clients actually sustain?

Monthly packs filed without a meeting, quarterly meetings run on the packs. More frequent reporting decays into noise; less lets drift and world-changes accumulate unseen. The rhythm also matches the tooling — stale checks, diffs and evidence all run naturally at those intervals.

How do we report an incident to a client?

From the log, fast, with the incident format this series uses: what happened, the counts that measure it, what changed, what verifies the fix, and what routine now prevents the class. Clients forgive incidents reported this way; what they do not forgive is learning about one from their own screenshot.

Build my block kit — free · Browse the agent database

Related guides

HONEST SCOPE: this policy blocks KNOWN agent user-agents and behaviors listed below. Sophisticated spoofing can bypass any user-agent block (e.g. an agent sending a stock browser UA, or rotating residential IPs). This is a control, not a guarantee.
← Back