The median AI citation lasts 18 days. If your refresh cadence is quarterly, you are watching half of your wins expire between cycles. If it is annual, you may as well not be running an AEO programme at all.

So what does a sustainable rhythm look like — one a small team can actually run for years without burning out. We have been running a 4-week version since 2025, on roughly 40 client portfolios. This is the shape that survived contact with reality.

Why a rhythm beats a calendar

Calendar-based refresh schedules — “we refresh all evergreen pages every March” — miss the curve. A page might be decaying right now in May. The March 2027 refresh will be too late.

A rhythm is different. Every week the team does a small slice of the same work. The portfolio is on rotation. The pages that need attention bubble up to the front of the rotation because they are the pages losing citations. Pages that hold position get less work because they need less.

The 4-week version below is opinionated. You can run a 2-week or 6-week rhythm with the same principles. We landed on 4 weeks because it matches the natural cadence of B2B content teams and the typical decay-detection window. If you are a high-velocity content shop (ecom, news), shorten it.

Week 1 — measurement only

The temptation in week 1 is to start editing. Resist it. The whole week is for looking, not touching.

Citation-status pull. For every tracked prompt across every engine, log the citation status today. Compare to last week’s pull. Flag pages that lost a citation. Flag prompts where a competitor entered the cited set.

Decay-risk scoring. For each tracked page, compute a simple risk score — days since last edit + days since last cited + position in the candidate pool. The top 5 highest-risk pages become the working set for the next 3 weeks.

Opportunity log. New prompts that emerged this month — from GSC, from client conversations, from competitor tracking. Add them to the prompt universe. Pick one or two to target with new content over the quarter.

Outputs. A one-page status doc — what is cited, what is decaying, what is new. Three numbers: total active citations, citations lost this week, new prompts entering tracking. The doc is the input for weeks 2–4.

Time budget. 2–3 hours for a 30-page portfolio. Mostly automated with the right dashboard — the human part is interpretation, not data entry.

Weeks 2–3 — micro-edits

This is where most of the value lives. Small, repeated edits to the top decay-risk pages. Not full rewrites. Targeted moves that signal “this page is being maintained.”

Date stamps. Bump dateModified on pages that got real edits. Add a visible “Updated May 2026” line in the page header. Both signals matter to the retriever and we have measured them adding 40–60% to evergreen half-life.

Refreshed examples. Each top page gets one example swapped for a current one. The 2024 case study becomes the 2026 case study. The “last quarter” stat becomes “this quarter.” Small change, big freshness signal.

‘What changed’ section. A 3-bullet block at the top of any page that had material change. “Updated: added 2026 data, refreshed pricing table, new section on entity disambiguation.” LLMs reward this. Humans reward it more.

Quick Facts row addition. Every top page gets one new row in its Quick Facts table — a new data point, a new comparison, a new constraint. Each Quick Facts row is independently extractable, so a new row can unlock a new prompt on its own.

Schema spot-check. Validate the JSON-LD on the page using a schema validator. Catch the breakages that happen when CMS migrations or content edits silently delete schema blocks. Re-add what was lost.

Outputs. Top 5 decay-risk pages all get the 5 moves above. That is 25 micro-edits across 2 weeks. Roughly 4–8 hours per week of writer + light engineering time.

Common failure. Treating micro-edits as “low priority, do them when you have time.” They are not low priority. They are the highest-ROI hours in the whole rhythm because they extend the half-life of citations you have already earned.

Week 4 — one full refresh

One page. The single highest-decay-risk page from week 1’s scoring. Full refresh — body rewrite, new schema, fresh examples, updated internal links.

Body rewrite. Not a copy edit. A real reread of every section, asking “is this still the sharpest version of this claim, and is the answer to the dominant prompt still in the first paragraph.” Rewrite where the answer is no.

Schema upgrades. Add anything missing — Article, FAQPage, BreadcrumbList, sameAs, dateModified. Run the validator. Re-test in a search-result preview tool.

Internal links. Reroute internal links from the rewritten page to your current strongest related pages. Anchor text on those links matters — use brand + category patterns where you are linking to entity-anchored pages.

Author and E-E-A-T. Verify the byline still points at a named expert with a live LinkedIn. If the original author left the team, reassign or coauthor.

Outputs. One page goes from “decaying” to “fully refreshed and visibly maintained.” dateModified updated. Resubmitted to Bing/IndexNow. Logged in the rhythm tracker.

Time budget. 4–6 hours for a writer + 1–2 for an engineer. Plan it on a Friday so the page goes live ahead of the week 1 measurement of the next cycle.

Quarterly overlay — structural rework

Every 12 weeks, the rhythm runs once more — but with a quarterly tier of work running in parallel.

Schema upgrades. New schema types as the spec evolves. Each quarter we look at what the spec added, what AI engines started reading, and where we can add coverage. Always one schema tier above the previous quarter.

llms.txt revision. The file decays the same way pages do. Refresh the entries, prune dead URLs, add new hub pages, recheck the format against the current draft of the spec.

Sitemap audit. Catch orphan pages, dead pages getting indexed by accident, hreflang gaps. The sitemap is the retriever’s map — keep it clean.

One new pillar or hub. Each quarter the portfolio gains one new pillar page on a fresh prompt cluster. Not five new pages chasing five new prompts — one deep pillar that earns citations slowly across multiple adjacent prompts.

The quarterly tier is the strategic work. The weekly rhythm is the operational hygiene. Both have to run; one without the other does not move the score.

Owners and ownership

The rhythm works because each week has a clear primary owner.

Week 1 — Analyst. Pulls the data, scores the risk, hands the working set to the writer.

Weeks 2–3 — Writer. Owns the micro-edits. Has authority to ship date stamps, examples, Quick Facts rows, and schema spot-checks without escalation.

Week 4 — Editor and Engineer. Editor leads the body rewrite. Engineer handles schema, internal links, IndexNow resubmission, sitemap update.

Quarterly — Practice lead. Owns the strategic overlay. Decides which schema upgrades, which new pillar, which prompts to chase.

On a small team one person can wear multiple hats. The role labels matter less than the clarity about who is responsible for which output each week.

Common failure modes

We see four. All of them kill rhythms.

Skipping week 1. Editing without measurement means editing the wrong pages. The page that feels stale to you is rarely the page that is actually decaying in the data. Always measure first.

Treating the rhythm as project work. Project work has a finish line. The rhythm does not. Month 12 looks identical in shape to month 1. Teams that treat AEO as a project end up in calendar-based mode by month 4.

Letting micro-edits slip. “We had a busy week, we will catch up next week.” Two weeks of skipped micro-edits and citation count drops 15%. We have watched it happen on every client portfolio at least once.

No quarterly tier. Running only the weekly rhythm without the structural quarterly work means the portfolio’s ceiling never rises. The micro-edits hold the line on existing pages, but the new prompt clusters and new schema work are what create new ceiling.

For the structural moves themselves — what schema to add, what llms.txt should contain, how to choose new pillars — see the AEO roadmap and the schema stack. The rhythm above is the operating system that runs those moves week after week.

The full rhythm is what one of our Growth-tier retainers buys — every week, every quarter, named owners, tracked output. If you would rather run it in-house, the structure above is the whole thing. The hardest part is not designing it. It is not skipping week 1.