Turn an Underperforming Page into a Content Refresh Brief

A page can lose search traffic while still containing the most useful explanation on your site. Rewriting everything risks deleting that value. Updating the date and adding a few paragraphs risks leaving the actual problem untouched.

A useful content refresh brief makes a narrower decision: what should change, what deserves to stay, and what evidence would justify the work? It gives an editor enough detail to improve the page without having to repeat the investigation.

Claude or ChatGPT can help assemble that brief from your performance exports, current copy, customer questions and RankParse page checks. The important division of labor is straightforward. Your exports establish performance. RankParse establishes selected facts about the public page. A person evaluates whether the proposed answer actually serves the reader.

The workflow below produces a ready-to-assign editorial document, including evidence, edits, assets, acceptance criteria and a measurement plan. All examples are illustrative, not results from a customer or a live research run.

Define the job before collecting more data

Choose one URL and one reader task. “Improve our inventory management article” leaves almost every decision open. “Help an operations manager decide whether a spreadsheet can support inventory across two warehouses” gives the editor a useful boundary.

Write the desired next action as well. It might be downloading a stock-transfer worksheet, comparing software options, or understanding when a spreadsheet is sufficient. A page that answers an early research question should not automatically be judged by immediate demo requests. Conversely, a buying guide that attracts irrelevant readers can gain traffic without becoming more useful to the business.

Prepare a small input packet:

• The target URL, intended audience, reader task and business purpose.
• Search Console page and query exports for two comparable periods, with filters and dates preserved.
• Relevant analytics or conversion exports, including the event definitions.
• Current article copy, screenshots of important interface examples and any downloadable assets.
• Recent customer questions, product documentation and claims that require a subject expert.
• Constraints such as keeping the URL, preserving a calculator or excluding unapproved product promises.

Remove personal information from customer notes. Summaries of recurring objections usually provide enough context. If the page has already been chosen through a broader prioritization exercise, carry its evidence into this packet instead of starting another site-wide audit. The Search Console opportunity workflow explains how to organize those exports.

Inspect the page that exists today

Connect RankParse through the MCP setup guide. The Claude and ChatGPT OAuth connector uses https://connect.rankparse.com/mcp and provides read-only research tools. For this workflow, upload your own Search Console and analytics exports; the web connector does not directly supply them or edit the article.

Ask for get_page_seo with the complete public url. Its documented output includes title.text, description.text, headings.h1, headings.h2, canonical.url, robots.raw, status_code and final_url. These fields help establish whether the intended page is accessible and whether its visible promise is reflected in the server-delivered structure.

Two limitations matter to editorial work. The tool parses raw HTML rather than executing the page's JavaScript, and headings.h2 contains the first ten H2 headings. An absent section in that list does not prove the complete article lacks it. Supply the full copy and inspect the rendered page before describing a missing answer as a confirmed gap.

The word_count field is also an observation, not a target. A short page may solve its reader's task completely. A long page may bury the answer. Similarly, title.optimal and description.optimal report length heuristics; they do not assess whether the wording is persuasive or whether it matches search intent.

Use the first prompt to establish evidence without rushing into a rewrite:

We are refreshing this one URL: [public URL].
Audience: [role and situation]. Reader task: [specific decision].
I have attached current copy, two dated Search Console exports,
conversion definitions and customer questions.

Use RankParse get_page_seo for this URL. Build an evidence ledger:
observation, source, date, limitation and editorial implication.
Separate observed page facts from hypotheses about performance.
Check status_code, final_url, canonical.url and robots.raw first.
Do not write replacement copy yet. Flag missing inputs explicitly.

If the returned title belongs to a bot challenge or an error template, stop interpreting the audit as article evidence. Inspect the page manually and record that the fetch was inconclusive. If final_url differs from the chosen URL, establish which page owns the brief before comparing performance against the wrong document.

Turn query changes into reader questions

Search Console queries help describe what people asked, but a table of queries is not yet an editorial plan. Group them by the decision the reader needs to make. Preserve useful distinctions: “inventory spreadsheet template,” “spreadsheet inventory limits,” and “inventory software migration” can overlap while requiring different answers.

For each group, compare clicks and impressions across the supplied periods. Consider position and click-through rate in context rather than treating an average as a complete diagnosis. A page can appear for a broader set of queries and show a lower average position without losing its established audience. Query exports can also be incomplete, so query totals need not equal page totals.

Have the assistant describe alternative explanations. Fewer impressions may reflect changed demand, changed coverage or visibility loss. Stable impressions with fewer clicks may invite inspection of the page's promise and the current search results. A public RankParse page check cannot determine which explanation caused the decline.

If search intent is disputed, provide dated observations from manually reviewed results. Record the query, country, device and result types you saw. Neither a page audit nor a competitor's heading list is a substitute for that context. Avoid telling the assistant to copy whatever competitors have added; their pages may serve different customers.

Worked example: evidence becomes a decision

Consider a fictional article called “Inventory Spreadsheets for Growing Teams.” The figures below are illustrative inputs, deliberately simple enough to follow. They are not RankParse benchmarks or measured customer results.

The resulting instruction is not “make the article longer.” It is: keep the template, explain its limits for two warehouses, add a worked transfer example, and give the reader a clear point at which software becomes worth evaluating.

That decision also defines what stays out. A broad history of inventory accounting would distract from the task. A detailed ERP implementation guide belongs elsewhere. A generic FAQ repeating the introduction would add little value. The brief should give the editor permission to exclude these additions.

Write changes as assignments with acceptance criteria

A refresh brief becomes practical when each recommendation identifies the exact section, proposed change, supporting evidence and completion check. “Improve expertise” is difficult to assign. “Have the operations specialist review the transfer example and confirm how in-transit stock is counted” has a clear owner and result.

Ask the assistant to classify existing material as keep, revise, add or remove. Require a reason for removals, especially when a section attracts links, supports conversions or contains unique expertise. Your supplied copy and a manual review should settle whether a passage is redundant; an abbreviated heading output cannot do that alone.

The second prompt converts reviewed evidence into an editorial plan:

Use the reviewed evidence ledger to draft a refresh brief.
Keep the existing URL and working download. Organize changes as
keep, revise, add or remove. For each change give the section,
reader question, evidence reference, exact assignment and acceptance
criterion. Mark claims requiring expert review.

Separate necessary corrections from optional experiments.
Do not invent customer quotes, product capabilities or research.
Include an explicit list of useful material we must preserve.

For the fictional inventory page, an acceptance criterion might require the example to reconcile opening stock, transfers, receipts and closing stock across both warehouses. Another might require the introduction to state who the spreadsheet suits before the first download link. These checks evaluate usefulness instead of demanding an arbitrary number of words or headings.

Use RankParse's links.internal and links.external only as counts. They cannot identify which internal destinations need updating. Supply a crawl export with source and destination URLs, or inspect relevant links manually. The internal linking workflow can help turn that evidence into specific contextual links.

Give the editor a complete handoff

Copy this structure into your project system. The headings are brief fields, not a required structure for the finished article.

Refresh brief
Target URL:
Audience and reader situation:
Primary question and desired next action:
Reason for refresh, with evidence references:
Performance periods and filters:
Current-page check date and limitations:

Preserve:
Required corrections:
Section-level editorial assignments:
New example or downloadable asset:
Product and factual claims requiring review:
Internal links, with verified source and destination:
Proposed title and description, with rationale:
Explicit exclusions:

Editor:
Subject expert:
Technical owner:
Acceptance criteria:
Publication and tracking checks:
Measurement date and comparison method:
Unresolved questions:

For each new asset, explain why text alone is insufficient. A stock-transfer worksheet lets readers test their own process; a decorative diagram may add no additional understanding. Specify who supplies the underlying numbers and checks the calculation. If no reliable example is available, make obtaining one a dependency instead of asking the model to invent operational evidence.

Title and description proposals should reflect the updated page accurately. Keep multiple options only when they represent distinct, defensible promises. The description.optimal field can flag a length issue, but your reviewer must decide whether the description names the actual benefit and avoids unsupported outcomes.

Review the brief before anyone rewrites the page

A convincing brief can still contain unsupported causal claims. Ask the assistant to challenge its own assignments using the original evidence ledger. The final prompt should focus on review rather than generating another alternative document:

Review this brief against the supplied evidence and page checks.
Identify assignments that depend on unverified assumptions, conflict
with the reader task, remove proven utility or require unavailable
assets. For every blocker, name the smallest check that resolves it.
Return a final assignment list and a separate unresolved list.
Do not claim the refresh will recover a particular amount of traffic.

Resolve disagreements with the person who knows the audience. For example, sales may want a stronger demo invitation while support wants a clearer troubleshooting section. Place each request against the reader's stage and the evidence for the page. A brief should make that trade-off visible rather than silently adding both sections.

When evidence is weak, narrow the change. Correcting obsolete product instructions may be justified immediately, while reframing the entire article around a different audience may need more research. Record those decisions so an editor does not interpret every brainstormed suggestion as mandatory scope.

Verify publication and measure the intended change

After publication, run get_page_seo again and compare the specific fields affected by the work. Confirm the final URL, title, description, canonical declaration and robots meta. Check the rendered example, download, mobile layout and conversion event separately. A correct metadata response does not prove that a worksheet opens or that a table is usable on a phone.

Preserve the publication date and a change log. Compare the same query groups and audience segments used in the brief after enough time and observations have accumulated for the page's traffic level. Look at whether the intended reader finds and uses the new answer, not only at total clicks. Record other changes that could affect interpretation, including campaigns, seasonality and major site releases.

If the page improves, avoid attributing every gain to the refresh without qualification. If it does not, return to the explicit hypothesis: was the missing transfer explanation actually the problem, did the page reach the right audience, and did the implementation satisfy the acceptance criteria?

Start with one underperforming page, collect the small evidence packet, and use RankParse to establish its current public state. The finished brief should let an editor make a precise improvement and let you later judge whether that improvement helped.