Keep data content fresh with an AI agent
At Ahrefs, data-driven articles are core to content marketing. Posts like "Top Google Searches" or "Most Asked Questions on Google" attract significant organic traffic – but only when the numbers are current. Google visibly rewards freshness: after each update, clicks and visibility often rise sharply. That is why someone had to maintain these posts regularly – a process that used to consume entire afternoons monthly or quarterly.
Si Quan Ong from the Ahrefs content team describes how the team automated this recurring workload with an AI agent in Letaido. The result: at least 20 saved hours per month, monthly instead of quarterly updates, and room for more data-based formats – without sacrificing editorial quality control.
Why freshness matters for data-driven SEO content
Data articles live on credibility. Anyone publishing search volumes, questions, or rankings must ensure readers and search engines see current values. Google recognizes stale content and favors pages that receive substantive regular updates. For publishers, each refresh can stabilize or expand rankings – provided the numbers are accurate and tables remain readable.
That sounds manageable with one post. With 14 or more data-driven articles, the process becomes a burden: pull data from Keywords Explorer or the API, clean tables, trigger design for custom charts, paste everything into WordPress, adjust dates, and republish. A single afternoon can pass without any strategic new work.
From quarterly compromise to agent workflow
Ahrefs long updated data-driven posts only quarterly; some formats were never touched. The bottleneck was not motivation but repetitive craft. With Letaido – Ahrefs' agentic platform – an automated "Data Refresh Hub" has been running for two months. It fetches fresh datasets monthly for 14 defined sources, cleans them by fixed rules, and creates WordPress drafts with new tables. The agent then emails that a human can review.
The build was less spectacular than effective. US keyword tables ran through Letaido, global datasets through an internal service, AI citation tables through Brand Radar – each with its own interfaces. Even small issues like a field name "Cpc" instead of "CPC" cost time. Crucially, the agent was built conversationally in Letaido, not by hand-coding – including error fixing.
What the agent handles – and what stays human by design
The Data Refresh Hub automates data retrieval, cleaning, formatting, and draft creation. It does not rewrite articles or publish on its own. Instead it delivers traceable intermediate results: what was removed, what was kept – visible next to the final tables.
Data cleaning requires judgment
For "Most Asked Questions on Google," sorting by search volume is not enough. The raw list contains movie titles, brand searches, and pseudo-questions. For people rankings, an LLM layer checks up to 5,000 candidates for real names versus irrelevant terms. Colleague Louise built an agent for fastest-growing companies – with rules against word double meanings ("Cursor," "Perplexity"), one-month spikes, and wrong brand matches. Humans make such decisions; the agent implements them at scale.
Ryan Law, Director of Content Marketing at Ahrefs, uses the same logic for his own data formats: scheduled monthly runs, fresh charts and tables, WordPress drafts, date and sample-size adjustments – all prepared, live only after human approval. Three team members now run similar workflows without central mandate.
| Task | Agent | Human |
|---|---|---|
| Fetch and format data | Automated | Define rules |
| Clean lists | LLM + rules | Review results |
| Publishing | Draft only | Approval and publish |
Technically, the hub draws on multiple Ahrefs data sources: Keywords Explorer for search volumes and questions, Brand Radar for AI citation tables, and internal services for globally assembled datasets. That mix explains why implementation was complex despite seemingly simple table updates – and why agentic orchestration makes more sense here than a single script.
Adoption spread without management directive. Si Quan Ong shared the workflow internally, and colleagues adapted the principle for their own rankings and charts. Ryan Law describes the effect as real automation rather than demo: hours of tedious routine work disappear while SEO-relevant content stays noticeably fresher for readers and search engines.
How content teams find automatable SEO routines
Recurring marketing tasks can often be identified when two questions are answered yes: Does the task run on a fixed rhythm, and can quality criteria be clearly described? "Fetch the same metrics from the same source monthly and paste them into tables" passes both. "Rewrite the article" usually does not – and that part often should stay with the editor.
- Automate the mechanics: retrieval, cleaning, formatting – not editorial judgment.
- Make filter decisions transparent so errors are visible before publish.
- Allow drafts only; a human publishes and takes responsibility.
- Lock critical metrics and headline stats so the agent cannot quietly change verified numbers.
Gartner predicts more than 40 percent of agentic AI projects will be scrapped by end of 2027 – often because demos promise more than productive workflows deliver. The Ahrefs approach works because it is boring and narrowly defined: it does not replace writing but the thankless maintenance of data-driven SEO content. For teams with many number-heavy posts, that can be the difference between rare updates and reliable monthly freshness.
