Guides / DNSFilter

An AI usage report per client, from DNSFilter

DNSFilter is MSP-native: sites, roaming agents and per-client organizations under one console. That structure makes it the cleanest source in this series for producing separate, client-labeled AI audits at scale.

Layer: DNS
Identity: site, roaming agent, network
Format: CSV query log export
Built for: MSP multi-tenant workflows
Run a free audit on a DNSFilter export
The workflow

One console, three clients, three reports

The core discipline: one export per client organization, one upload per client, one labeled report per client. Never merge tenants.

Export per organization

Pull each client's query log CSV for the same 30-day window. Same window across clients is what makes your cross-client benchmarks honest later.

Upload with the client label

Run each file separately and enter the client's name at run time. The label prints on every page of that client's PDF, which is what makes it a deliverable instead of an internal artifact.

Deliver and book the repeat

PDF to the client with your recommendations, CSV into your PSA notes, next quarter's run into the calendar. The commercial framing lives on the MSP page.

Separate runs are also the data-handling answer: each client's evidence, retention and deletion stay independent, which is the sentence your MSA reviewer wants to read.

The export

What a DNSFilter query log gives the audit

The query log CSV carries the domain, the requesting identity and the policy result. That is the whole recipe.

timestamp,domain,agent,site,action "2026-09-15T08:55:12Z","chatgpt.com","LT-ACC-014","Accounting Office","allowed" "2026-09-15T08:57:40Z","grammarly.com","LT-ACC-014","Accounting Office","allowed" "2026-09-15T09:01:23Z","leonardo.ai","DT-MKT-03","Main Office","allowed" "2026-09-15T09:03:58Z","carbonads.net","DT-MKT-03","Main Office","allowed" "2026-09-15T09:06:32Z","otter.ai","Roam-Sales-07","Roaming","allowed" "2026-09-15T09:09:47Z","deepseek.com","DT-ENG-11","Main Office","allowed"

Agents and sites become sources in the report. The roaming sales laptop transcribing calls off-site is the classic line-five finding.

Fields the audit reads

  • domain: the hostname to match, the one required column.
  • agent / site: identity at device or location grain, whichever your deployment logs.
  • action: allowed vs blocked, kept for enforcement evidence.
  • timestamp: window verification and clean re-runs.

Grain expectations

  • Roaming agents give device-level attribution, the best DNSFilter offers.
  • Site-only deployments attribute per location, which still answers "which client office does this".
Worked scenario

Three clients, one Tuesday morning

An eight-person MSP runs the same 30-day export for three very different clients. Sample data; the spread is the lesson.

Accounting firm, 35 staff run 1 of 3

9 AI tools. Chatbots plus a document summarizer, nothing abusive.

Finding: the summarizer's training terms are unstated, and tax documents are the workload.

Action: enterprise alternative sanctioned, consumer domain blocked, client signs the quarterly cadence.

Law office, 20 staff run 2 of 3

6 AI tools, but one is a dictation app on two roaming laptops, terms silent on training.

Finding: privileged audio potentially processed by an unvetted vendor.

Action: immediate block plus a vetted legal-transcription tool; the report PDF goes to the managing partner same day.

E-commerce brand, 60 staff run 3 of 3

22 AI tools, heavy image and copy generation, two flagged training-by-default.

Finding: product photography edited through consumer tiers, client IP in the uploads.

Action: team plan migrations, and the MSP's first paid "AI governance" retainer conversation.

Same export shape, three different risk stories. That spread is why per-client reports beat one merged view: each PDF speaks that client's language to that client's board.

Cross-client reading

The benchmark layer only MSPs have

After a quarter of consistent windows, your client base becomes its own reference data. Three comparisons clients pay attention to:

Tools per 10 staff

The e-commerce brand's 3.7 versus the law office's 3.0 reads differently once clients know the peer range. Nobody wants to be the outlier without a reason.

Training-exposure share

The percentage of tools that train on data by default or opt-out. A client above your book's median gets the terms-review pitch with evidence attached.

Sanctioned coverage

Clients with a real approved list trend toward high sanctioned share within two quarters. The laggards are your governance-retainer pipeline.

Output

What each client's report contains

Report sectionWith a DNSFilter CSV
Header and scopeThe client's name as entered at run time, on every page.
Summary tilesFull totals per client: tools, high-risk, training exposure, abusive, unsanctioned.
Tool tableEvery matched domain with category, risk, sovereignty and dated training verdicts.
Per-source tableAgents and sites with tool counts and hits: which office, which roaming laptop.
Sanctioned splitThat client's approved list against their observed traffic.
CSV + PDFThe PDF is the deliverable; the CSV feeds your PSA and their risk register.

Each upload is read once and discarded. Reports live 90 days in your account and can be deleted per client, earlier on request. Put those two sentences in the engagement letter.

DNSFilter edges

Details that matter in managed deployments

Threat categories are not AI inventories

  • DNSFilter's categorization protects; it does not enumerate per-tool risk or training terms.
  • The audit's dated verdicts are the layer your client's compliance answer needs on top.

Roaming agents extend the frame

  • Agent-equipped laptops log DNS off-network, so home-office AI use appears attributed to the device.
  • Clients hear "coverage follows the agent", which is also an upsell for agent rollout.

Guest and IoT segments

  • Site-level identities lump guests into the location. High AI counts on a guest network are a policy note, not a personnel one.

Volume planning

  • SMB clients run small: a 30-day query log for a 50-person client typically fits a full report with room.
  • Caps: 2,000,000 lines or 25 MB per run; bigger single clients go through contact.

The exact PDF your client would receive

Open the sample evidence pack with the client-label header, the tool table and the per-source breakdown. Attach it to your next QBR agenda and let it sell the assessment.

Open the sample report
Economics

What three clients a quarter costs you

The scenario's morning used three full audits. Ways to cover that, from lightest to standing.

3-pack, $199

$66 a report, no subscription. Right for testing the motion on three friendly clients, exactly like the scenario.

5-pack, $299

$60 a report. An engagement season's inventory for a small book of clients.

MSP plan, 10 a month

The standing allowance with client labels, for when the quarterly cadence covers your whole base. Full detail on pricing and the MSP page.

FAQ

DNSFilter export questions

Can I audit all clients in one merged file?

Technically yes, practically no. Merged tenants produce one blended report, which is useless as a client deliverable and messy as evidence. One export per client.

Which DNSFilter export do I pull?

The query log CSV per client organization over your window, keeping domain, agent or site, action and timestamp columns.

Do roaming agents change the report?

They improve it: device-level attribution and off-network coverage. Site-only clients still get full tool inventories with location-grain sources.

How do the client labels work?

You enter the client's name when you run the audit; it prints on every page of that PDF. Labels are per run, so three runs mean three cleanly branded deliverables.

Can clients see each other's data?

No. Each report is its own document from its own upload, in your account. Separate uploads keep evidence and deletion independent per client.

What if a client asks how detection works?

Hostnames from their own logs, matched against a register of 20,399 classified AI domains maintained daily. The methodology page is client-shareable.

Delivery kit

What accompanies the PDF when you deliver it

The MSPs who convert audits into retainers wrap the report in three small artifacts of their own.

The one-page cover letter

Your logo, three findings in plain language, three recommended actions with effort estimates. The client forwards this page, not the PDF, so it carries your brand.

The decision table

The report's block/control/allow verdicts as a checklist the client can approve line by line in one meeting. Approved lines become your tickets.

The delta line

From the second quarter on: "tools 22 to 17, training exposure 8 to 3, sanctioned share 20% to 74%". One sentence that renews the engagement by itself.

Consistency rules

Five habits that keep multi-client audits comparable

Benchmarks and trend lines die from sloppy inputs. These five habits keep the data honest across a client book.

Fix the window

  • Calendar month, every client, every quarter. Mixed windows make every comparison an argument.

Fix the columns

  • Export the same four columns each time. A client whose export gains or loses the agent column shifts attribution grain mid-series.

Label identically

  • Same client label spelling every run, or your own archive becomes ambiguous by year two.

Log the run in the PSA

  • Date, window, report link, verdict counts. The next engineer to touch the account inherits context, not folklore.

Fifth habit: read each report against the client's own sanctioned list, not a generic one. Building that list is covered on sanctioned vs unsanctioned AI.

Three deliverables by lunch

Pick three clients, pull three query logs, run three audits. The Tuesday-morning scenario is a repeatable play, not a story.

Start with a free audit