Claude Prompts: Long-Context and Analysis Templates
XML-tagged prompt skeletons built for Claude's long context and its strength at following structured instructions.
<role> You are a senior analyst reviewing the material inside <documents>. You are direct, evidence-driven and you flag uncertainty instead of hiding it. </role> <task> [WHAT TO PRODUCE — one sentence, verb first] </task> <documents> <doc id="1" title="[TITLE]"> [PASTE DOCUMENT 1] </doc> <doc id="2" title="[TITLE]"> [PASTE DOCUMENT 2] </doc> </documents> <constraints> - Cite every claim with the doc id and a short quote. - If the documents disagree, show both positions rather than picking one silently. - If a claim appears in no document, label it "outside the provided material". - Do not summarise documents one by one; synthesise across them. </constraints> <output_format> 1. ANSWER — max 5 sentences. 2. EVIDENCE TABLE — claim | doc id | quote | confidence (high/medium/low). 3. CONTRADICTIONS — where documents disagree and which is better supported. 4. GAPS — what the documents do not cover. 5. NEXT ACTION — one concrete step. </output_format>
Long context is not infinite attention. If the answer misses something, move that document to the start.
02 — Variables
Fill in the brackets
The highlighted fields are the places where your own details go. You do not have to fill all of them; anything left empty is completed by the model's own assumption.
[TITLE]
[PASTE DOCUMENT 1]
[PASTE DOCUMENT 2]
03 — How to use
How to use it, step by step
- 01
Use XML tags consistently — Claude follows them more reliably than markdown headers in long prompts.
- 02
Give each document an id; citations become checkable instead of vague.
- 03
Put the constraints after the documents, not before, when the documents are long.
04 — Variations
Same prompt, different result
One prompt does not solve every case. These variations ask for the same job in a different tone, length or format — pick the one that fits.
Contract review
Same structure for contracts: extract obligations, deadlines, termination conditions, and flag clauses that conflict across documents.
Codebase review
Same structure for code: each file as a doc, output a risk table with file, line, issue and severity.
05 — Tips
Notes that take the result up a level
- Claude is strongest when the task is "read all of this and tell me what matters" — long context plus synthesis.
- Ask for citations with document ids; without them you cannot verify a synthesis across 40 pages.
06 — FAQ
Frequently asked about ChatGPT Prompts
How much can Claude read at once?
Very large contexts, but quality degrades toward the middle of very long inputs. Put the most important document first and last, and ask for per-document citations.
Do XML tags actually matter?
They help on long, multi-part prompts by giving the model unambiguous boundaries. On short prompts they add noise.
Next
You may also find these useful
ChatGPT Prompt for Summarising Long Documents (3 Layers)
Turn a 30-page report into something readable in ten minutes: one-sentence core, an executive summary and a bulleted detail layer.
Beginner ChatGPTClaudeChatGPT Idea Generation Prompt: 30 Ideas, 5 Angles
Ends the "nothing comes to mind" loop: multiplies one topic across five angles and scores every idea for feasibility.
Beginner ChatGPTClaudeChatGPT Persona Prompt: Expert Role Template
Instead of a shallow answer to the same question, get an expert's view: identity, experience, bias, boundaries and output rules.
Beginner ChatGPTClaudeTranslation Prompt That Keeps Tone and Terminology
Instead of a literal translation, keep meaning, tone and technical terms — and make the model check its own work at the end.
IntermediateThis page was prepared by the CopyPromt editorial team, first published 25 October 2026. Prompts are written to work across models; brand names are mentioned for identification only.