Ahrefs' Data Refresh Hub pulls fresh data for 12 datasets every month, then stops and waits for a human to review everything before anything publishes. That pause is the difference between a refresh workflow you can trust and one that quietly rewrites 30 stats posts overnight with numbers nobody checked. Refreshing existing posts is also a far narrower job than "generate articles at scale," which makes it a safer place to start.
This guide shows how to automate content refreshes for data-driven and evergreen posts without tying you to one vendor. You will audit which posts qualify, write the process down, pin approved data sources, set accept/reject/escalate rules, and keep a human approval step before anything goes live.
The human-review-before-publish pattern comes from Ahrefs' guide to automated content creation. Its author describes the Data Refresh Hub, which also rebuilds tables and charts, updates WordPress drafts, and emails him when it finishes. This article has not run or tested that workflow. It borrows the structure and adds its own scorecard, templates, and checks.
Which refresh tasks are safe to automate, and which need a human?
Split the job by judgment, not by effort. Anything that follows a fixed rule and can be checked against a source is a candidate. Anything that requires deciding what the numbers mean stays with a person.
Safe to automate (as a draft, never a live publish):
- Pulling figures from a named source
- Filtering rows by written rules
- Rebuilding tables and charts
- Formatting numbers consistently
- Producing a before/after diff
Needs a human:
- Deciding whether a changed number alters the post's argument
- Rewriting the intro or headings
- Changing the title
- Handling results that fit no rule
- Clicking publish
The Ahrefs guide's own manual process includes a judgment step: the author checks whether changes are big enough to warrant rewriting the intro or subheadings. That is a good reason to keep this step out of the automated path.
Step 1: Audit which posts qualify
Start with one task, not a platform. The Ahrefs guide suggests looking for work that is repetitive, time-consuming, governed by reasonably stable rules, valuable when done correctly, and frequently delayed or skipped. Turn those five traits into a scorecard.
Score each post group from 0 to 2 on each trait, for a maximum of 10. Use this cutoff, which is a judgment call rather than a standard:
- 8 or more: candidate for automation
- 5 to 7: partial workflow
- Below 5: stay manual, with a checklist
Illustrative example: a 30-post blog
This is a made-up scenario to show the method. The scores are not measurements from a real site.
| Post group | Posts | Repetitive | Time-consuming | Stable rules | Valuable | Often delayed | Total | |---|---|---|---|---|---|---|---| | Industry statistics pages | 8 | 2 | 2 | 2 | 2 | 2 | 10 | | Pricing comparison pages | 6 | 1 | 2 | 1 | 2 | 2 | 8 | | Evergreen how-to guides | 10 | 0 | 1 | 1 | 1 | 1 | 4 | | Opinion and case studies | 6 | 0 | 1 | 0 | 1 | 0 | 2 |
The result: automate the eight statistics pages first. Pricing pages come second, once you trust the first workflow. Give how-to guides a manual checklist and leave opinion pieces alone. One narrow job beats a sprawling one. See more on how to prepare shorts for ask youtube: a creator audit for additional background.
Step 2: Write the process down
The Ahrefs guide makes a blunt point: an agent cannot reliably automate a process nobody can explain. Most people perform repeat tasks on autopilot, so reconstruct what you actually do, including shortcuts and corrections. Do not rely on what an old SOP says.
An AI assistant can help surface unwritten rules by interviewing you. The guide offers its own version of this idea. Here is an original prompt you can paste and adapt:
I want to automate refreshing [post type, e.g. statistics pages]. Interview me about how I do it today, one question at a time. Start with where the numbers come from and end with how I publish. If I say "usually," "it depends," or "I just know," ask what the exception is and when it applies. Ask what I check before I trust a number. At the end, produce a numbered checklist that lists every rule, every exception, and every point where I make a judgment call. Mark each judgment call as ESCALATE.
Save the output as a single document your tool reads on every run. If your tool supports saved instructions, use that feature; the Ahrefs guide calls its version a skill file. Whatever your tool calls it, write the process once so you never have to re-explain it.
Step 3: Pin approved data sources
This step prevents invented numbers. The Ahrefs guide warns that language models make up figures and that open web search can surface an old listicle or low-grade content. Its rule of thumb: the model's job is to process the data, not to be the data.
For each post group, write a source register with one row per dataset. Name the exact source, how the workflow accesses it (API, export file, or manual download), and what the workflow may do with it. A minimal version looks like this:
Dataset: [e.g. average product price table]
Approved source: [exact URL, API endpoint, or file]
Access method: [API / CSV upload / manual paste]
Allowed operations: [rank, round, filter, reformat]
Forbidden: [estimating, filling gaps from memory, using any other source]
If source is unavailable: stop and flag, do not substitute
The last line matters most. A workflow that quietly substitutes a different source when the approved one fails lets slop in. A failed source should halt the run and notify a person.
Step 4: Add accept, reject, and escalate rules
Instructions help, but examples help more. The Ahrefs guide recommends showing the system what you would accept, what you would reject, why, and which ambiguous cases should go to a human. It also warns that a simple rule, such as keeping every query that starts with a question word, is not as simple as it sounds. Test your rules against messy rows, not tidy ones.
Use a three-outcome template for every rule:
RULE: [plain-language rule]
ACCEPT example: [real row that passes] | Why: [reason]
REJECT example: [real row that fails] | Why: [reason]
ESCALATE example: [ambiguous row] | Why: a person must decide
DEFAULT when unsure: ESCALATE
Set the default to escalate, because a tool that must guess will guess confidently. Also write down the date rule explicitly. As a recommendation, change the visible "last updated" date only when figures or text actually changed, never as a cosmetic edit.
Step 5: Keep a human approval gate
Build the workflow so its final output is a draft, a diff, and a notification, never a published page. This mirrors the Ahrefs pattern, where the system emails the author and waits for review before anything is published.
Also read: go deeper on growth leader first 30 days: find vs believe channel audit
If your CMS allows it, give the automation account permission to create or edit drafts but not to publish. That turns the human gate into a technical control instead of a good intention. Name one person responsible for sign-off on each run.
Step 6: Verify before anything goes live
Run this checklist on every refreshed post. For the first few runs, check every page. Once the workflow has proven itself, move to a sample, but keep a full check for any post where the numbers changed direction.
- Data source match. Confirm every figure traces to a source in the register. Reject any number with no listed origin.
- Number spot-check. Open the original source and compare a set of figures to the draft. Include the top row, the bottom row, and any figure that changed sharply. Check rounding against the rule you wrote.
- Diff review. Read the full before/after diff, not just the new text. Look for deleted sentences, altered headings, changed links, and edits outside the intended scope.
- Rule audit. Review every row flagged ESCALATE, plus a handful that the workflow accepted and rejected, to confirm the rules behaved as written.
- Narrative check. Ask whether the intro, headings, and conclusions still match the new numbers. If a trend reversed, a person rewrites the copy.
- Before/after sample. View a few refreshed pages as a visitor would, on desktop and mobile, to catch broken tables, charts, and images.
- Human sign-off. The named reviewer approves, records the date, and publishes manually.
Limitations
Fresher dates do not guarantee rankings. The Ahrefs guide says Google's Query Deserves Freshness system boosts recently updated pages for topics that need current information, and that AI assistants prefer citing fresher content. Those claims come from a vendor with a product to sell, and this article has not verified them. Treat freshness as one possible factor, and update because stale numbers mislead readers.
Tool-specific claims deserve the same caution. The guide's author chose Letaido, a platform from his own company, and describes its connectors and data access. You can run the same steps on other agents or automation tools, but the guide notes that you may have to connect data sources, a CMS, or databases yourself. It does not cover costs, setup effort, or reliability.
Edge cases still need people. Sources change formats, datasets get discontinued, and a changed number can undermine a post's whole premise. No checklist anticipates everything, which is why the default is escalation. Finally, this article offers no results or traffic figures, because none were measured.
What to expect next
Your first run will probably produce more escalations than you expected, and that is useful. Each one points to a rule you never wrote down. Add it to the checklist, rerun on the same post group, and tighten the source register. Extend the workflow to the next group on your scorecard only after the first group runs cleanly through the human gate.
