Use the FIRST framework, a worked example, and a downloadable review record to check flow, identity, reality, safety, and terms without repeating the launch checklist.

Website client onboarding should not end when the client receives a login and attends a training call. After a week, the site has met real requests, real schedules, and at least a few moments when nobody from the agency was guiding the next click.

That makes the week-one review different from launch quality assurance and different from the initial handoff. You are not rerunning every acceptance test. You are checking whether the operating arrangement works: the journeys reach the right place, the right people have the right access, early evidence is being collected, changes remain reviewable, and support requests have a clear commercial route.

I use five review areas: Flow, Identity, Reality, Safety, and Terms. Together they form FIRST. The name is useful because this is the first operational review, not a declaration that the site is permanently solved.

The week-one rule

Review evidence from real operation, then make one decision for every finding: fix it, teach it, reassign it, scope it, or accept it. Missing evidence is Unknown, not Green.

You can follow the framework in this article or download the week-one website client onboarding review record.

What the week-one review is for

The initial website handoff checklist answers whether the client has appropriate access, can complete an everyday edit, understands the boundary, and knows how to get help. The week-one review asks what happened after that controlled exercise.

Did the client try the task again? Did a form reach the intended person? Did someone request a change that crossed the agreed boundary? Did the client know whether an AI-generated update was still a draft? Did analytics collect data in the property the client actually owns?

This is also why I would schedule the review before the handoff call ends. A calendar event creates a real transition point between guided training and ordinary operation. Without one, onboarding often dissolves into scattered support messages.

Three related checks with different decisions
MomentPrimary questionEvidenceDecision
LaunchIs this identified release ready for visitors?Acceptance tests, production checks, approvalsShip or hold
HandoffCan the client operate the agreed scope?Real account, first edit, ownership and recovery notesTransfer or continue training
Week oneDoes the operating arrangement work without constant guidance?Actual requests, journeys, access, changes and support recordsClose, correct, extend, or move into ongoing care

A week is not enough time to prove search growth, conversion performance, accessibility conformance, or a return on investment. It is enough time to discover whether the site and the service are producing usable evidence.

Run the FIRST website client onboarding review

Give each area one status:

  • Green: evidence matches the agreed operating state.
  • Yellow: the site is usable, but a correction or clarification has an owner.
  • Red: a material journey, access, safety, or ownership problem needs immediate action.
  • Unknown: nobody collected the evidence yet.

Unknown matters. A form that nobody tested is not Green because no complaint arrived. An analytics dashboard with no traffic is not proof that tracking failed or that visitors did nothing. The review should preserve the difference between observation and assumption.

Flow: check the real journey from entry to destination

Start with the few journeys the website exists to support. For a service business, that may be a phone call, a quote request, and a booking. For a lead magnet, it may be submission, result delivery, consent capture, and human follow-up.

Run each critical journey on the production hostname with clearly labeled test data. Do not stop when the browser shows a success message. Confirm the notification, record, booking, payment, or other intended destination. If a request crosses into another system, that destination needs its own evidence.

I would also review the first week's genuine messages with private details removed from the discussion record. Look for operational questions:

  • Did the submission include enough context for the client to act?
  • Did it reach a monitored person rather than an old agency address?
  • Did the visitor receive the expected confirmation?
  • Did the phone, map, calendar, or email link behave correctly on a phone?
  • Did an error state explain what to do next?

The output is not “forms checked.” Write the actual route and result: “Homepage consultation form submitted at 10:14, confirmation displayed, record appeared in the approved CRM pipeline, notification reached the sales inbox, and the client assigned it.”

If the flow fails, separate a product defect from a process gap. A missing CRM record may require a technical repair. A record that arrived but sat unassigned may require a client owner. The visible symptom can be similar while the next action is different.

Identity: retest ownership and access through real accounts

An invitation is not the same as working access. Ask the client to sign in using the account they will actually use, complete one agreed task, and locate the support or recovery route. Then test one action that should remain restricted or require review.

For Fork projects, the current roles and permissions documentation distinguishes built-in and custom project access and warns that project access does not automatically grant sandbox administration. That is a product control. “Please do not change navigation” is an operating instruction. It does not become an enforced permission because it appears in the handoff notes.

Review ownership beyond the website editor. Name the current owner for the domain, platform account, billing, analytics, Search Console, form destination, CRM, and any connected service. If a client employee left or a contractor finished, review that person's access rather than assuming offboarding happened elsewhere.

Google's Search Console permissions guide is a useful example of why ownership needs precision. Owners and users have different rights, a property needs a verified owner for users to retain access, and a removed verified owner's token may need separate review. That does not mean deleting every unfamiliar token on sight. Some verification methods can support other services. It means the client needs a known owner who understands the current setup.

If the website has customer accounts, test those application permissions separately. A platform role determines who can operate the project. It does not automatically determine which customer can see a particular record inside the application.

Reality: read early evidence without inventing a story

The first week is good for verifying instrumentation and collecting questions. It is usually poor evidence for a sweeping performance conclusion.

Google says its Analytics Realtime report can be used to confirm that data is being collected, while other reports may take 24 to 48 hours to populate. That supports a narrow test: perform a labeled visit and action, then confirm the intended property receives it. It does not prove consent handling is correct, every event is configured well, or the week's conversion rate is meaningful.

Review search setup with the same restraint. Google's Search Console starting guide recommends verifying ownership, checking whether Google can find and read pages, considering sitemap submission, and monitoring performance. For a week-one review, I would check ownership and obvious crawl or indexing issues. I would not promise rankings from seven days of data.

The most useful reality evidence may be the client's own attempts. Ask three questions:

  1. What did you try to change or check?
  2. Where did you hesitate or stop?
  3. What did you send back to the agency instead?

Record the answers in the client's words. Then classify each one as a training gap, setup problem, defect, access issue, recurring care task, or new request. This turns “the client seems confused” into something an agency can improve.

Safety: inspect changes and the recovery route

List every production change made since handoff, including changes made by the client, the agency, and an AI agent. For each one, record the request, surface changed, release owner, checks performed, and any follow-up.

A useful AI request includes both an outcome and a verification target. Fork's Agent documentation gives the example of asking for a responsive settings page and verification at phone and desktop widths rather than merely asking to improve CSS. During the review, compare the client's requests with that pattern. Improve the next request before building a long prompt library.

Recheck the affected page, not the entire internet. At minimum, inspect the changed content at representative phone and desktop widths, follow affected links, and test the relevant form or interaction. If the client added an image or a new section, spot-check the title, headings, alternative text, keyboard route, visible focus, and form labels where relevant.

W3C's Easy Checks can help with a first review of titles, images, headings, contrast, resizing, keyboard access, focus, and forms. W3C is explicit that these quick checks are not a definitive accessibility evaluation. Label the work honestly.

Then confirm the recovery route. Fork checkpoints file-changing agent work and provides history and snapshot tools, but restore replaces the current workspace state. The week-one review should identify who can recover, which version would be used, and what later work must be preserved first. A website snapshot should not be described as reversing form deliveries or records in external systems.

Finally, assign ongoing review ownership. W3C's guidance on ongoing accessibility monitoring emphasizes named responsible people, clear scope and frequency, follow-up procedures, and a process for checking changed pages. The same operating discipline helps with more than accessibility: a check without an owner becomes a suggestion.

Terms: turn findings into service decisions

The review is incomplete until every material finding has an owner and a commercial route.

Route the finding before debating the solution
Finding typeTypical evidenceNext route
DefectApproved journey does not meet acceptance criteriaRepair under the agreed launch or warranty terms
Training gapClient has access but cannot repeat the agreed taskSimplify the guide and repeat one real exercise
Access issueWrong role, owner, recipient, or recovery contactAuthorized administrator corrects and retests
Recurring careTask will repeat and needs monitoring or reviewAdd it to the maintenance scope
New requestUseful work outside the approved deliverableUse a bounded change request and estimate it

Do not hide new scope inside onboarding just because the request arrived quickly. Do not invoice a missed acceptance criterion as an improvement. The classification protects the client relationship and the agency's margin at the same time.

I would end the review with one of three decisions: close onboarding, continue onboarding with a dated follow-up, or move into a named ongoing service. “We will keep an eye on it” is not a fourth decision.

A worked example for a fictional design studio

Assume an agency launched a five-page website for a fictional architecture studio. The site has a project inquiry form, a phone link, a portfolio the client can update, Google Analytics, and Search Console. The client is allowed to edit project text and images. Navigation, forms, and integrations require agency review. This is an illustrative setup, not a customer or a report of results.

During the first week, the client adds one portfolio image, asks the AI editor to shorten a project description, receives two genuine inquiries, and messages the agency because they cannot find one Analytics report.

Illustrative week-one decision record
FIRST areaEvidenceStatusDecision
FlowLabeled test inquiry and two genuine inquiries reached the approved inbox; phone link worked on a representative phoneGreenKeep the route and record the monitored owner
IdentityMarketing lead could edit a portfolio item but could not change navigation; former photographer still had a collaborator accountYellowClient administrator reviews and removes unneeded access
RealityRealtime collection verified; Search Console ownership confirmed; report question remainedYellowTeach the one report tied to the studio's goal, not a dashboard tour
SafetyNew image looked correct on desktop but its crop obscured the subject on a narrow phoneYellowAgency corrects the crop and adds mobile image guidance
TermsClient requested a new project filter not included in launch scopeGreenCreate a separate change brief and estimate

The review does not produce a single score. It produces five decisions. The site can remain operational while the access list, image guidance, and report training are corrected. The new portfolio filter does not need to masquerade as an onboarding emergency.

Use a 45-minute review agenda

I would keep the meeting short enough to stay operational:

  1. Five minutes: restate the agreed client jobs and support window.
  2. Ten minutes: run the most important Flow with labeled test data.
  3. Ten minutes: let the client sign in and show what they attempted.
  4. Ten minutes: review changes, collection checks, open questions, and recovery ownership.
  5. Ten minutes: classify findings, assign owners and dates, and choose the next operating state.

Send the completed record after the meeting. A recording can help, but the action register is the durable output. Each item should say who owns it, when it is due, and what evidence will close it.

Do not overclaim from one week

A first-week review should not become a performance theater exercise. Avoid these conclusions:

  • “Traffic is low, so the website failed.” The site may be new, reports may be delayed, or acquisition work may not have started.
  • “No support tickets means onboarding succeeded.” The client may not have attempted the task.
  • “The automated scan passed, so the site is accessible.” A preliminary check has a narrower meaning.
  • “The page loaded, so the integration works.” The destination still needs evidence.
  • “The prompt says not to change it, so it is protected.” Instructions and permissions are different controls.
  • “There is a snapshot, so every action is reversible.” External records and later work may remain outside that snapshot.

The honest outcome can be Unknown. Unknown with a named test is more useful than Green without evidence.

Close website client onboarding with an operating decision

A good week-one review makes the client less dependent on improvisation and the agency less dependent on memory.

Check the real Flow. Verify Identity through the accounts in use. Read Reality without forcing a success story. Inspect Safety around actual changes. Turn Terms into named work. Then close onboarding, extend it deliberately, or move the agreed responsibilities into ongoing care.

If the client still cannot complete an agreed edit, redesign that route before adding more capability. Fork is built for agencies creating client websites with AI access, including a visible path from request to verified change. If that model fits your service, join the Fork waitlist and use the FIRST record to plan the review you will run after the first week.

Written by Chris Peterson. Sources checked 2026-11-03.