Skip to content
DeepShellDeepShell
Documents and data · Marketing teams

Extract every H2 from a competitor's article

Use selected browser sources to extract every H2 from a competitor's article. DeepShell assists with extract work; you prepare the inputs, verify the evidence, and complete any external action.

Process guideIllustrative — extension test pendingUpdated 2026-09-06

Before you start

Inputs: Selected article.

Define: the audience, date or version, comparison criteria, and intended destination.

Access: open the sources yourself and select only content you are authorised to send to a model provider. Inaccessible or unselected pages are outside the task.

The process, step by step

  1. 1
    You complete

    Prepare selected article. Confirm you can share the selected material with a model provider.

  2. 2
    You complete

    Open the relevant pages or attachments, select the sources in the panel, and set the scope. Log in or navigate to any required pages yourself.

  3. 3
    DeepShell assists

    Extract the requested fields from the selected material. Request: Heading order, Exact H2 text, Section link.

  4. 4
    You complete

    Reconcile row counts, headers, units, and totals with the source; inspect first, last, and unusual rows.

  5. 5
    You complete

    Copy or export the reviewed result, then update the destination system yourself if needed.

Requires another tool: sending, publishing, form submission, CRM updates, purchases, and schedules are outside this extension workflow. Use the destination application yourself.

Your reusable workflow pack

Replace bracketed fields before use. Keep the sources and output together for the next reviewer.

Reusable prompt
Task: Extract every H2 from a competitor's article.
Use only these selected sources: Selected article.
Audience: [add the intended reader].
Scope: [add dates, criteria, versions, or exclusions].
Return these fields or sections: Heading order, Exact H2 text, Section link.
Cite the source and page or section for factual statements. Mark missing information as "Not stated". Separate facts from assumptions. Do not invent figures, decisions, or events.
Finish with unresolved questions and a review checklist. Do not send, submit, purchase, or update another system.
Output template
Heading order: [fill from the selected source; otherwise Not stated]
Exact H2 text: [fill from the selected source; otherwise Not stated]
Section link: [fill from the selected source; otherwise Not stated]

Output structure

Marketing & SEOIllustrative workflow
Example output

Heading order / Exact H2 text / Section link

  1. 01Heading order: [source-grounded heading order]
  2. 02Exact H2 text: [source-grounded exact h2 text]
  3. 03Section link: [source-grounded section link]
Reusable output template. Bracketed fields are placeholders, not observed results. Replace them only with information supported by your selected sources.

Review, then complete the process

  • Reconcile row counts, headers, units, and totals with the source; inspect first, last, and unusual rows.
  • Every material claim has a source you can open and inspect.
  • Missing information is explicitly marked; no placeholders remain in the final deliverable.
  • A named person has approved the result before it is shared or used.

Copy or export the reviewed result, then update the destination system yourself if needed.

Done means: the intended reviewer can trace the result to its sources, all missing details are resolved or explicitly flagged, and any external action has been completed by an authorised person.

If the result is incomplete

Check source selection and access first. Narrow the question, specify the missing field, or supply an authorised source. For long or scanned documents, verify that the relevant text is readable. If evidence is still missing, record the gap rather than accepting an invented answer.

Time, cost, and validation

No measured run time, cost benchmark, or end-to-end extension test is recorded for this guide. Actual cost depends on source length, model, and retries. Use the settled receipt from a real run; do not treat the example as a performance promise.

Practical questions

What should I prepare?

Selected article. Define the audience, scope, and required fields before you start.

Which steps are automated?

Extract the requested fields from the selected material. Source selection, review, and any sending, submission, or system update remain your responsibility.

What if the source is incomplete?

Keep missing values marked as Not stated. Add an authorised source or narrow the task; do not ask the model to fill gaps with plausible facts.

Has this workflow been tested in the extension?

This is an illustrative process guide. No end-to-end extension test or measured cost benchmark is recorded for this guide.

Continue the work