How to pull approved language verbatim across documents

Before you start: This article assumes you've already linked your downstream document to an upstream source using data tags. If you haven't done that yet, read How to link documents and maintain a consistent narrative across your dossier first, it covers the full setup workflow. Come back here once your data tag is pointing to the right document and you're ready to write the instruction.

Writing verbatim instructions

The AI instruction block is where all verbatim control happens. A well-written one-line instruction is enough to tell the AI exactly what to reproduce and how faithfully to reproduce it.

Be specific about what to copy

A bare instruction like "copy verbatim" tells the AI to reproduce the entire document. Name the table, section, or passage you want:

• "Find the [table name] in [document name] in section and recreate this table exactly."
• "In [report name], find table [X] and recreate it exactly."
• "Find the schedule of activities table and recreate it exactly as in the source document."
• "Copy the conclusions verbatim from the referenced document. Do not paraphrase."
• "Use the exact phrasing from the Summary and Conclusions section of the referenced document."

The AI understands document names, table names, and plain-language descriptions — use whichever is most precise for your content.

Scope to a specific file

For maximum precision, scope the data tag to the individual source file your instruction references. You can do this two ways:

• In the data tag field beneath the instruction block, clear the category tag and select the specific file
• Or, inside the instruction block itself, use the @ button to tag the file directly — type @ and select the document from the dropdown. This anchors the instruction to that exact file and ensures the AI pulls from the right source if other files in the program are updated

Directing the AI to a specific section

The AI reads the whole upstream document by default. To narrow its focus, name the section in the instruction:

"Focus on the Summary and Conclusions section of the referenced document."

There is no section-level tag, specificity comes entirely from the prompt.

Pulling tables verbatim

For tables, a one-line prompt is all you need. Reference the table by name, number, or description:

"Find the [table name] in [document name] in section and recreate this table exactly."

Multi-page tables: If the table spans multiple pages in the source document, add a stitching instruction so the AI outputs it as a single table rather than separate tables for each page:

"Find the [table name] in [document name], stitch it together, and generate one table as the output."

Footnotes: To carry footnotes through, add the instruction explicitly:

"Include the footnotes exactly as they appear beneath the table in the source document."

Example 1: pulling a summary section into a Module 2.3 overview

Say you're writing a Quality Overall Summary (Module 2.3) and need to carry the conclusions from your already-approved Module 3.2.S drug substance section forward exactly — the language has been reviewed and the phrasing must not change.

In the instruction block, you would write:

"Find the Summary and Conclusions section in Section 3.2.S of the referenced document and copy it verbatim. Do not paraphrase or condense."

This instruction names the section so the AI doesn't pull from the full document, and it explicitly prohibits paraphrasing. After generating, open the Sources panel and confirm the citation points to your 3.2.S document — not the Data Room.

Example 2: pulling a tabular summary into a Module 2.4 overview

Say your Module 2.6.2 pharmacology summary contains a tabular summary that spans multiple pages across two tables. You need that table carried into Module 2.4 exactly as it appears: same structure, same data, same footnotes.

In the instruction block, you would write:

"Stitch together the tables in Section 2.6.2 Table 1 and Section 2.6.2 Table 2 and recreate it exactly. Include all footnotes exactly as they appear beneath the table in the source document."

After generating, open the Sources panel and confirm the citation points to your 2.6.2 document — not the Data Room.

Verifying the output

After generating, open the Sources panel and hover over the content. Confirm the AI cited the correct upstream document — not the Data Room. Then compare the output directly against the source to confirm the language or table was carried through as intended. The Sources panel tells you where the AI read from; the content check is yours.

Tips

1. Verbatim and synthesis can coexist. Different sections in the same document can use different instructions — some pulling verbatim, others synthesizing. Configure each section independently in Template View.
2. If you get gaps in a table — question marks or "data not shown for brevity" — the instruction is likely too broad or the data tag is pulling from more than one file. Tighten the table name in the instruction and scope the data tag to the specific source file. Use the three-dot menu on the section to regenerate that block alone without touching the rest of the document.
3. Verbatim works within the same document too. Set the data tag to Current Document to pull content from earlier in the same document. The same instruction patterns apply.

• Linking documents — Full setup workflow for data tags and upstream/downstream document logic. See How to link documents and maintain a consistent narrative across your dossier.
• Source Verification — After generating a verbatim section, use verification mode to confirm the AI read from the correct upstream document and flag any content it couldn't trace.
• AutoReview — Use AutoReview after generating interconnected sections to check that language and conclusions are consistent across the full dossier.