LLM visibility checklist

Objective

Test access, entity resolution, naming, citation and variation separately. The controlled asset is the AI visibility matrix. Completion means the firm can inspect the work, reproduce the check and continue without depending on an unexportable vendor view.

When to use this procedure

Run this SOP when crawl access, entity facts and test prompts needs a new baseline, a correction or a scheduled review. Do not begin because a dashboard has issued a generic warning. Write the decision the procedure is meant to support. That keeps the work bounded and gives the quality gate a real pass condition.

Assign one operator and one approver for the AI visibility matrix. The same person may fill both roles for a small firm, but the worksheet should still separate the act from the approval. Attorney review is required whenever published text could state law, characterize a charge, describe a defense or expose a client fact.

Inputs

Procedure

1. Open the control record

   Create the AI visibility matrix under firm ownership. Write the market, practice scope, responsible person and review date before changing anything.

2. Capture the starting state

   Inspect crawl access, entity facts and test prompts. Save the raw export or source copy and note any field that cannot be observed.

3. Set the acceptance rule

   Define what counts as complete, which value must remain fixed, and who can approve an exception. Put the rule in the worksheet.

4. Perform the work

   Test access before interpreting generated answers. Make one traceable change at a time when the system allows it, then log who acted and when.

5. Test the result

   Repeat the observation with the same settings. Compare the output with the acceptance rule and retain fetch records and answer samples.

6. Close or recover

   If the gate passes, prepare the handoff. If it fails, restore the last known state where possible and open a blocker record.

Asset-specific record

For LLM visibility checklist, the first inspection is crawl access, entity facts and test prompts. Save that state before acting and name the person who can confirm it. The working action is to test access before interpreting generated answers. Put that sentence in the task record so a later operator can distinguish the intended change from incidental edits made during the same session.

The proof file is fetch records and answer samples. Give it the procedure number and observation date, then store it with the source material used for the decision. If the platform has no export, capture the complete visible settings and account context. A screenshot can show a state; it cannot replace an available raw file used for a calculation.

Asset closeout

The deliverable is channel-specific finding list. Review it against the starting state and acceptance rule before transfer. The firm should receive an editable or exportable copy under its control, along with the next review date. Record any dependency that remains open and the person responsible for resolving it.

Stop the closeout when access failure makes answer testing meaningless. Preserve the failed state and write the expected result beside the actual one. If a repair changes the observation method, close the old version as inconclusive and open a dated procedure. That keeps the original baseline intelligible.

Quality gate for LLM visibility checklist

Approve the AI visibility matrix only when a second operator can reproduce crawl access, entity facts and test prompts and reach the same disposition. The reviewer must locate fetch records and answer samples, identify who approved it and state the limit that travels with the result. Return the LLM visibility checklist file if a conclusion depends on an unrecorded setting or a vendor-only screen.

Scenario test for LLM visibility checklist

Run a dry test on the AI visibility matrix before declaring this procedure ready. Begin with crawl access, entity facts and test prompts and give the test file to someone who did not perform the setup. That reviewer should be able to locate the starting record, identify its date and explain which field the procedure may change. If those answers depend on a verbal explanation from the operator, the LLM visibility checklist package is not ready for repeat use.

Next, ask the reviewer to follow this working instruction: test access before interpreting generated answers. Compare the resulting observation with fetch records and answer samples. A useful comparison names the unchanged settings as well as the changed field. It also records an absent value as absent; it does not fill a blank from memory or copy a later value back into the baseline. File the review beside the AI visibility matrix so the test travels with the work.

Acceptance example

A passing LLM visibility checklist file produces channel-specific finding list that another firm-authorized operator can open and continue. The package identifies the approval rule, links to fetch records and answer samples and gives the next review a calendar date. The receiving operator can tell which source controls the fact and which worksheet contains the observation. No private case detail is needed to understand the result.

A failed example begins when access failure makes answer testing meaningless. Leave that state visible in the AI visibility matrix record. Assign the repair to the person who controls the source, then repeat crawl access, entity facts and test prompts under the recorded settings. Do not replace the failed copy with the repaired one. The two files show what changed, and they let the approver decide whether the original acceptance rule still fits the LLM visibility checklist procedure.