Give the developer a reproducible website symptom and a way to verify the intended behaviour. An assistant citing an old page can start an investigation. The developer needs the address it cited, what happens when someone opens it, which destination the business approves and what evidence would show the route has been fixed.
Keep the assistant observation, the website diagnosis and the proposed change in separate fields. They may be related, but they come from different checks. That separation gives the developer a bounded task and gives marketing a way to follow up without keeping an otherwise complete release open indefinitely.
AI Visibility is published by Lunasol, which offers GEO and AEO services. This guide uses a public agency case, Google documentation and a fictional developer ticket. Sources were checked on 5 November 2026. AI Visibility publication disclosure
Record what the customer could see
Start by saving the relevant part of the observation in the approved project record. Include the exact question if it matters, the date, the experience being used and the link or factual statement that prompted the investigation. A developer can use a source link to reproduce a route. A cropped answer card may leave the destination ambiguous.
Name the experience precisely. Google’s official 13 October 2026 model announcement describes multiple products and subscription contexts in one caption. That illustrates why recording a model name alone leaves the observation underspecified. When relevant, record whether you used a consumer interface, an API or a separate web search mode. Note the conditions you actually observed. Google’s official model announcement
Explain the customer consequence in a sentence. A cited page might route a visitor to the wrong branch or contain a booking instruction that has expired. A missing citation might warrant investigation, but the absence of a link gives the developer less to reproduce. Start by checking the intended public source before opening a change request.
If the observed answer has no source link, preserve that fact. Compare its wording with the business’s approved information and inspect the relevant pages independently. If you find a reproducible problem there, describe that problem on its own. The ticket can be justified by the effect on the public page while the source of the assistant’s wording remains unknown.
Reproduce the website issue on its own
Open the exact address and note the starting environment, whether you are signed in, and the path taken. Record the final destination after redirects. If the issue is intermittent, include dates and the conditions under which it appeared. For a landing page issue, check whether it is specific to one address, a language path, a device or a wider template.
Ask the developer which layer should be tested. A successful server response can contain only an application shell. A browser may then load the content with JavaScript. Google describes crawling, rendering and indexing as separate stages and explains that it can use rendered HTML. Its troubleshooting guide recommends checking rendered content when diagnosing JavaScript issues. Record the representation you inspected before saying the content is missing. Google JavaScript SEO basics and Google JavaScript troubleshooting
A ticket can say that a required field appears in an ordinary browser but was absent from a particular inspection result. Include the result and the date. The developer can then investigate how that field is populated, whether a required resource failed and whether the issue affects other pages. Avoid prescribing a rebuild based on the first screenshot.
If the brief contains a technical hypothesis, label it. A note saying “likely shared routing rule” lets the developer check the pattern. A note saying “change the routing rule” assumes the investigation already established the cause. Leave room for a smaller fix if the issue comes from an individual link or an obsolete record.
Choose a diagnostic that answers the question
Google’s URL Inspection tool distinguishes the indexed view from a live check. Its documentation says the live test follows redirects without displaying the final address it tested. For a redirect issue, retain a separate route trace and inspect the intended destination directly when needed. A live result is a dated diagnostic of access and parsing under that tool’s conditions. Google URL Inspection documentation
Choose a test that matches the expected behaviour. If the issue is that the old URL leads to the wrong destination, record the initial address, response and final address. If the final page opens but a necessary service fact is missing, check the page content. If the developer fixes a shared component, verify representative pages that use it and a relevant page outside the intended change.
Ask who can supply private tool evidence. The website owner might provide the indexed observation or run a live test, while the developer checks the routing or rendering separately. Record the limits of each check in the ticket. The existing AI Visibility audit guide covers broader evidence collection. The brief starts when there is a specific task someone can investigate or implement. AI Visibility audit guide
Write a ticket a developer can estimate
The following example is fictional. A visitor follows a link to a retired booking page. The path redirects to a generic landing page, while the business has approved a specific replacement for the same service and language. The issue is the route a visitor can reproduce.
| Ticket field | Example entry |
|---|---|
| Customer impact | A visitor arriving at the retired booking path has to locate the intended service again |
| Reproduction | Open the retired path in a signed-out browser and record the redirect chain and final destination |
| Expected behaviour | The retired path should resolve to the approved replacement for the same service and language |
| Scope | Correct the named route and its approved equivalent if it is affected |
| Exclusions | Other migration routes and the booking form itself remain separate tasks |
| Acceptance | Each in-scope path reaches the approved destination and the destination displays the intended booking route |
| Regression | Check the equivalent path in the other language and a neighbouring service route |
| Follow-up | Analyst records a later answer observation as a separate dated check |
Attach the exact URLs, the date of the captured result and the internal approval for the new destination in a real ticket. If the retired page has no equivalent, ask the content or service owner to decide what users should see. The developer can explain what is technically possible, while the business chooses the intended customer journey.
A public 009 Agency case illustrates a division of work in which the agency says it supplied detailed technical specifications and the client’s developers deployed website changes. In your brief, say whether the developer is expected to investigate, implement or review a change prepared elsewhere. Ask who supplies the release evidence when those responsibilities sit with different teams. 009 Agency’s public INC4 case
Split a routing defect from a missing fact
Suppose the developer follows the retired path and reaches the correct replacement, but the replacement uses the wrong pickup instructions for one branch. The routing acceptance check now passes. The next task depends on where the instructions originate. If the business approved the wrong text, send it to the service owner. If the correct text was approved but a shared template displays a different branch’s instructions, give the developer that comparison.
A useful evidence pair would show the approved branch record and the live page, with their location identifiers and dates. Add a known working branch that uses the same template. The developer can compare the records and determine whether the issue is tied to the template, the individual content entry or the relationship between them. If the approved record is missing, the work can proceed as an investigation while operations confirms what should be displayed.
Write one acceptance check for the corrected branch and another for a branch that should retain its existing instructions. Ask the website team to confirm the change applies to the intended locations. A check of the affected page alone can miss a fix that makes every branch display the same text. If the template also serves another language, the reviewer should know whether that version uses the same data or has its own approved source.
Close the original route ticket if it already meets its conditions. Link the new content or template task to it so the programme owner can see why the customer experience still needs attention. This keeps a single observation from becoming an open-ended instruction to fix every related page.
Include what must still work after the change
The acceptance check should name the page, action and expected result. Add a small number of regression checks based on what the developer changes. A route fix might affect multiple languages or booking paths. A template fix might affect both the target pages and pages for other services. Check the shared boundary before choosing the sample.
Keep the test narrow enough to run before and after release. If a booking destination opens, say how far the reviewer should go. Opening a form can verify the route. A complete submission is a different check that needs an approved test arrangement and an owner who can confirm receipt. Use the business’s existing process for that verification.
Record which version reached production. Link the release or deployment record, note the test time and attach the result. If the test fails, specify whether the owner should correct the new version, restore the previous behaviour or hold the release. The team should be able to tell which condition triggered that decision.
For a shared change, let the developer identify which routes inherit it. If a fix can affect other pages, give the reviewer examples from both sides of the boundary. Record which pages were checked so a passing sample is understood in context. If one page behaves differently, capture the difference before deciding whether it belongs in the same ticket.
Close implementation and schedule the observation
A developer can close a route change when the agreed route behaves correctly and the checks have passed. The analyst can later observe whether a search product uses the new destination or still surfaces the old one. Keep that observation in a separate record with the date, question and surface. A new answer can inform the next investigation.
If a public result stays stale, compare its observed source with the current live page and the available indexing evidence. The result may point to another page or reflect an older version. Decide which new evidence would justify reopening the website ticket. If the site still behaves as specified, a separate monitoring task may be enough.
For an issue that cannot yet be reproduced, write an investigation ticket with a clear stop condition. It might ask someone to inspect two versions of a page, identify which one was seen and return the result to the programme owner. If the investigation finds a site defect, create the implementation task with the observed evidence. If it finds a missing business decision, assign the decision to the appropriate owner.
Before handing the work over, ask the developer to restate the expected behaviour and the reviewer to describe the evidence they will accept. A brief is ready when both people can point to the same destination and explain what the check will establish.



