Content refresh with search data: briefing AI to update what is already ranking

A new-content brief starts from a topic gap. You identify a keyword cluster your site doesn't cover, research the SERP, and tell the AI what to build. The inputs are competitive analysis, search intent, and a target cluster.

A refresh brief starts from a specific page that exists, has history, and is either losing ground or failing to reach its potential. The inputs are fundamentally different: position data over time, click-through changes, query drift from Search Console, and a read on what competing pages now do better. The goal is improvement within a working structure, not replacement.

When you hand AI a new-content brief, starting from scratch is appropriate. When you hand AI a refresh brief without the underlying data, it treats the existing page the same way — it rewrites rather than improves, discarding the structure and phrasing that search engines have indexed and trust. The brief determines whether the model helps you or costs you.

What the signals actually tell you

Three data signals drive a useful refresh brief. Position drift shows where a page has slipped. A page that ranked 4 now ranks 12. That gap usually means competing content has improved, not that yours was wrong. The brief should specify where the rankings fell and what the competing pages do differently in those sections.

Query expansion shows new queries the page is beginning to attract that weren't part of the original targeting. If an article on email marketing benchmarks now pulls queries about AI-generated subject lines, there is an emerging intent the current content doesn't address. That is a gap the refresh should fill, not ignore.

Traffic decay without position decay points to a CTR problem: the page holds position but fewer people click. This is usually a title or meta description issue, not a structural content problem. The brief should call that out explicitly so the AI addresses the right element.

Each signal maps to a specific brief instruction. The content maintenance features in Brass-SEO track these at the page level so you are not manually pulling individual GSC reports. Without the data, you are guessing at what needs to change. With it, the brief writes itself — the same premise that makes using search data to brief AI effective across all content types.

Building the brief from the data

The 30-minute GSC refresh process covers the data pull in detail. Once you have the signals, the brief has four components.

What to preserve names the sections that are performing. If the H2 structure is clean and the page holds a featured snippet, say so explicitly. The AI should work within that skeleton, not replace it. This instruction sounds obvious but it prevents the model from pruning something that is working.

What has changed specifies the drift with actual numbers. "Position for [primary keyword] dropped from 5 to 14 over six months" is useful. "Rankings dropped" is not. Pair the position data with what you know about the competing pages — new sections they have added, different angles they have taken, depth they have added to a topic your page covers thinly.

What to add lists the specific gaps: queries the current content does not address, sections that are thin relative to competing pages, claims that have aged out of accuracy. The AI needs this as a directive, not a general instruction to make the page better.

What must not change flags anything generating backlinks or conversions. If a specific section is the reason a page earns editorial links, the brief should say so. The AI will preserve it.

A brief structured this way produces a surgical edit. That is the point of a refresh brief — if the right answer were a full rewrite, you would not be refreshing.

Where AI earns its place and where it does not

AI handles section expansion well. If your page covers a topic in two paragraphs and the competing page covers it in five, giving the AI the existing section plus a directive to expand it — with context about what competing pages include — produces usable material quickly. The same applies to restructuring for clarity: reorganizing subheadings, improving transitions, adding an FAQ block are mechanical improvements that do not require deep domain judgment.

What requires human judgment is accuracy. Any claim that was true when the page was written may not be true now. Statistics age. Products change. Regulations shift. The AI will not flag these unless you explicitly tell it to — and even then, it can only flag what you have pointed at. Before publishing a refresh, someone needs to read every factual claim against the current source. For guidance on where that editing line sits, see our notes on editing AI-generated content.

The six-month re-optimization cycle is a practical cadence for deciding when to run this process. Most pages that warrant a refresh surface within that window. Pages that need refreshing more frequently usually signal a problem with the original brief, not with the content itself.

Frequently Asked Questions

How do I know when a page needs a refresh versus a rewrite versus deletion?

Refresh when the page holds some position and has backlink equity worth preserving — the structure is sound but specific sections have drifted. Rewrite when the original intent or structure is wrong and the position loss is significant. Delete when the page targets a topic the site no longer covers and is drawing the wrong traffic with no realistic path to relevance.

How long should a refresh brief be?

Long enough to cover the four elements above — what to preserve, what changed, what to add, what the page must not lose — with specific data attached to each. For most pages that is 300 to 500 words. Briefs that run longer usually indicate the content needs a rewrite, not a refresh. The content audit and refresh workflow covers how to batch this decision across a large page inventory.

How do I handle accuracy-sensitive claims in an AI refresh?

Flag them in the brief explicitly: "The statistics in the second section are from 2022 — verify current figures before outputting." The model will note the uncertainty in its output rather than fabricate updates. After the refresh is generated, a human still needs to check every flagged claim against the actual source. There is no shortcut for this step.

Does the refresh brief process change for long-form content?

The structure is the same, but the data layer expands. Long-form pages often rank for dozens of query variants, so the query expansion analysis carries more weight. Segment the brief by section — assign specific gap-filling directives to each H2 block rather than treating the page as a single unit to improve. This keeps the AI's edits targeted and prevents it from homogenizing sections that are already performing.