Check AI summary coverage before trusting the appendix

For AI research workflow, the first decision is whether one small example can be reviewed by a person. An AI summary can leave you unable to tell whether the appendix informed its conclusions. To check coverage, compare the report’s contents with the summary, then trace its claims to specific source pages. Record the body, tables, and appendices separately. Anything without a checked connection stays unverified.

Map the document: list its sections, tables, notes, and appendices with page locations.
Check the summary: connect each material claim to supporting text or a readable table.
Keep the gaps: record missing, inaccessible, and unverified material alongside the evidence.

This is a playbook for reviewing a publicly accessible report using a document viewer and a plain worksheet. It does not require a paid audit service. Access to a summarization tool is a separate condition.

Prepared: 2026-09-09. Tested date: not available. No report, summary, or completed comparison was supplied for this article. The source map below is a reusable artifact, not a test result. There is no verified appendix-coverage finding to report.

The question is what the summary represents

“Did the AI read the appendix?” sounds like a yes-or-no question. A summary alone does not establish the system’s internal reading process.

The useful research question is narrower: Does the summary accurately represent the appendix material that matters to this decision?

A matching phrase shows a connection. A checked citation supports a particular claim. Neither establishes that every surrounding page was considered.

Keep these observations separate:

• Present in the source: the material exists in the document.
• Represented in the summary: a summary passage reflects that material.
• Verified against the source: a reviewer checked the connection and its qualifications.

An appendix can be present without appearing in the summary. It can also supply a crucial limitation without deserving its own summary paragraph. The goal is faithful compression, not equal space for every heading.

A verified passage supports a claim; it does not certify the rest of the document.

Start with a stable report and a preserved summary

Choose a publicly accessible report whose body and appendices you can open. Prefer visible page labels and a usable contents page. These are selection criteria, not evidence that any particular file has been checked.

Record the document title, source location, edition or revision label, and your review date. Preserve the summary exactly as received.

Keep the original and any revised summary separate. Otherwise, a correction can erase the evidence of what was originally missing.

Check the page numbering before creating references. A viewer’s file position may differ from the number printed on the page. Record both where necessary, especially when front matter or appendices use different numbering.

If the appendix is a separate download, give it its own source entry. A link from the main report does not establish that the linked material was included in the summary.

Let the contents page define the expected scope

Copy the report’s section headings into your worksheet, then inspect the document beyond that list.

Look for tables, footnotes, methodological notes, and supplementary files. Add relevant items even if the contents page does not name them.

Give the body, tables, and appendices separate entries. A body paragraph might describe a result while a table note defines which cases it includes. Combining them into a single “results covered” row would hide that distinction.

For each entry, record:

• The exact heading or item label.
• Its printed page location and viewer position.
• Whether it is readable in your viewing setup.
• Why it could matter to the summary’s intended use.

That last field keeps the review focused. A methods appendix may explain exclusions. A supporting schedule may contain background detail. Decide relevance from the material itself, rather than from the word “appendix.”

Build the source map around checkable connections

Use the following worksheet as the audit artifact. Every bracketed cell is an instruction to fill in, not an observed result.

Artifact caption: An unfilled source-map template separating page evidence, summary connections, and unresolved scope. It shows the review structure, not demonstrated coverage.

Use a controlled set of statuses:

• Verified: inspected evidence supports the summary passage with its relevant conditions.
• Partial: a connection exists, but a material qualification is missing.
• Absent: no corresponding summary passage was found.
• Contradicted: inspected evidence conflicts with the summary claim.
• Unverified: the connection has not been checked or the source cannot be inspected.

“Absent” is not automatically an error. Explain whether the omission changes the reader’s understanding.

Check in both directions

Start with the summary. Take each material claim and locate supporting evidence in the report.

Record the page and the smallest passage that makes the connection checkable. For a table, include the row label, column label, unit, and relevant note. A number detached from those labels is not sufficient evidence.

Then reverse direction. Start from the source map and look for each decision-relevant source item in the summary.

This second pass asks a different question. Claim checking tests whether included statements are supported. Source-to-summary checking tests whether important material was left out.

For an appendix, identify whether it limits, explains, or changes a body conclusion. Follow that relationship into the summary. If the summary retains the conclusion but drops a necessary condition, mark the connection partial.

Check what the summary says, then check what the source needed it to say.

Keep uncertainty visible when the review gets awkward

No failed trial is documented here. The following are review conditions to watch for, not reported incidents.

A page is visible but its table is unreadable. Record the access problem. Do not infer the cells from nearby prose.

A citation points to the right section but misses the qualification. Keep the connection partial until the claim and its conditions agree.

An appendix finding also appears in the body. The summary may represent the finding, but that does not establish direct appendix use. Record the evidence location you actually checked.

Only selected pages were reviewed. Name those pages and retain the remaining scope as unverified. Do not label the whole report covered.

A clean worksheet should make these limits easier to notice, not easier to hide.

Finish with a decision the evidence can support

The final decision should concern the summary’s suitability for its stated use, not an unsupported verdict about internal document processing.

Use this handoff checklist:

• Document version and original summary are preserved.
• Body, table, and appendix entries remain distinguishable.
• Material claims have checked page evidence.
• Relevant qualifications survive compression.
• Missing and unreadable material remains visible.
• The conclusion names the review’s actual scope.

For this article, the decision remains unassessed because no comparison evidence was supplied. In a completed review, release the source map alongside the summary so another reader can inspect its basis.

Copy the source-map template and complete it against your report before relying on the summary.

Related build logs