Choose pages by Search Console impressions, not by age: start with articles that show 500 or more impressions in 90 days but sit between positions 8 and 20. Use AI to flag stale facts and thin sections, edit on the same URL, then compare each refreshed page's clicks and position over 28 days against untouched pages from the same client.
The tracking half is where most agencies fall down. They refresh twenty articles, rankings move, and nobody can say whether the refresh caused it. The fix is cheap: log the edit date, keep a handful of comparable pages unedited as a control group, and judge both over identical windows.
Choosing which old articles are worth the hours
A refresh budget is finite, so the first job is ranking candidates, and impressions are the best single signal because they show that Google already considers the page relevant to something. Export the Performance report for each client by page for the last 90 days, with clicks, impressions, click-through rate and average position. Paste the rows into a sheet and add three columns: article age, last edit date and the query with the most impressions.
Then sort into four buckets. This is the rule I would hand a junior on day one:
| Bucket | Signal in the export | Action |
|---|---|---|
| Near the top | Average position 8 to 20, impressions above 500 | Refresh first; small gains move many clicks |
| Seen, not clicked | Position under 8 but click-through well below the client's norm | Rewrite title, description and intro; leave the body |
| Faded | Impressions down 30% or more against the previous 90 days | Full refresh; check whether a competitor overtook it |
| Dead weight | Under 20 impressions in 90 days, older than 18 months | Merge into a stronger page or retire with a redirect |
The thresholds are starting points. A pest-control firm with a small local site may never see 500 impressions on any page, so scale the number to the client: take the top fifth of pages by impressions and apply the position rule inside that set. Write the chosen thresholds on the client's tracking sheet so a colleague can rerun the selection in six months and get a comparable list.
A worked selection for a car repair garage
Say a garage client has 64 articles. The 90-day export shows 11 pages with more than 400 impressions. Of those, four sit at positions 9 to 17: a tutorial on reading a dashboard warning light, a cost guide for brake pad replacement, a page on how often to service a car, and a seasonal tyre guide. Two more sit at position 3 with click-through of 1.2% when the garage's average is 4%. That makes six pages for this round: four full refreshes and two title-and-intro rewrites. Eleven hours of editing for six pages is far easier to defend than a blanket "update everything" retainer.
Also pick control pages now, before anyone edits anything. From the pages that qualified but did not make the cut (say the fifth to the ninth in the ranked list), set aside four and leave them untouched for the test window. They share the client's domain, seasonality and general authority, which makes them a fairer comparison than the site as a whole.
The Complete Excel Course with AI Integration
From formulas and pivot tables to Power Query, DAX, VBA and AI with Copilot, ChatGPT and Claude — hands-on from the first lesson.
By Sagnik Bhattacharya and Virginia Thorn, Primetime Emmy Award winner
Buy nowUsing AI to audit an article before you change a word
An audit prompt does more for quality than a rewrite prompt. Give the AI the article text, the top five queries from Search Console, and the date, and ask for a list of problems with the evidence for each. Keep it from rewriting: you want a diagnosis a human can approve line by line.
You are auditing an article for a refresh. Today's date is [date].
Article text is below. The queries that already bring impressions are:
[query 1], [query 2], [query 3], [query 4], [query 5].
Return a table with four columns: Section, Problem, Evidence, Suggested change.
Check for: statements that could be out of date (prices, product names, rules,
years), sections that do not answer any listed query, vague claims with no
number or example, missing steps a reader would need, and any heading that
repeats another. Do not rewrite the article. Do not invent sources.
If you are not sure a fact is out of date, say "verify" and name what to check.
[paste article]Here is the kind of output to expect, and what I would fix. This is an illustrative extract for the garage's brake pad guide:
| Section | Problem | Suggested change |
|---|---|---|
| "What it costs" | Quotes a price range that carries a 2023 date | Verify against the garage's current price list; replace with a range and the date checked |
| "Signs your pads are worn" | Four signs listed, no mention of the dashboard warning | Add the warning light and link to the warning-light tutorial |
| Intro | Does not answer "how long do brake pads last" in the first 60 words | Add a direct answer with a mileage range the garage will stand behind |
What you fix by hand: the AI cannot know the garage's own price list, so the price row stays a "verify" until someone looks it up. It also tends to flag perfectly good sections as thin because it has no sense of what the client's customers actually ask at the front desk. Ask the client's service adviser for three real questions and add them as input. That single step is what stops refreshes reading like every other refresh.
What to change, and what to leave exactly as it is
Refreshing is not rewriting. The text that earns impressions today should survive the edit, because Google has already decided it answers something. Before you touch the page, save three things in the tracking sheet: the title tag, the first 150 words, and the H2 list. If the page drops, you need a way back.
Work through the page in this order, and stop as soon as the page is good enough, because each extra change adds risk.
- Fix wrong or stale facts. Prices, product names, versions, rules, years in headings. This is the highest-value change and the one most worth doing by hand.
- Answer the main query earlier. If the top query is "how often should a car be serviced", the answer belongs in the opening paragraph, not four screens down.
- Fill gaps the audit found. A missing step, a missing comparison, a missing example with numbers. Ask AI for a first draft of the new section, then have the client or a specialist check it.
- Cut or merge duplicated sections. Two H2s that say the same thing split the page's focus.
- Improve the packaging. A table for a comparison, a short checklist, a clearer title. Keep the target phrase in the title.
- Update internal links. Link to the client's newer pages, and add links to this page from them.
A before-and-after for one section
Illustrative. The furniture maker's care guide for oiled oak tables had this paragraph before the refresh: "Oil your table regularly to keep it looking great. Most people oil theirs a couple of times a year." The audit flagged it as vague, with no method and no reason to trust it. After the refresh, with details supplied by the maker's workshop manager, it read: "Re-oil an oak dining table when water stops beading on the surface, which is usually every six to twelve months in a busy household. Wipe the top with a barely damp cloth, apply a thin coat of hardwax oil along the grain, leave it for 20 minutes, then buff off the excess. Skip the buffing and the surface stays tacky for days."
Notice that the AI wrote none of the facts. It restructured the paragraph once the facts had been supplied. That is the right division of labour: people supply the knowledge, AI tidies the prose and spots the gaps.
Facts that need a human, every time
Treat any figure, legal or safety statement, product claim or named tool as unverified until a person has checked it against a primary source. Keep a "facts checked" column on the sheet with the source and the date, which also gives you something to show the client when they ask what the retainer covered. For topics where being wrong has consequences, such as warranty terms or chemical safety on a cleaning company's site, send the changed paragraph to the client for sign-off before it goes live. The source-checking tutorial for agencies goes into the method.
Keeping the URL and handling dates honestly
Keep the URL. A refreshed page on its existing address keeps its links, history and the impressions you selected it for. Changing the slug to include a new year, for instance, forces redirects and discards the signal you were trying to improve. If a client insists on a new year in the headline, put it in the title and heading but leave the address alone.
Dates need more care. Search engines and readers both read them, and a date that changes without substance behind it reads as a trick. Google's own self-assessment questions for helpful content ask whether you are changing page dates to make them seem fresh when the content has not substantially changed. A sensible policy for every client:
- Show the original publication date and a separate "last updated" date on the page.
- Change "last updated" only after a substantive edit: facts corrected, a section added or replaced, advice changed.
- Keep the dates in the article's structured data (
datePublishedanddateModified) consistent with what is visible, as Google's guidance on byline dates asks. Validate the markup after each edit; the schema validation tutorial shows how. - Add a one-line change note at the bottom for big edits, such as "Updated with current brake pad price ranges and a new section on ceramic pads."
Request a recrawl with the URL Inspection tool in Search Console once the page is live, then log the date you did so. Google sets a quota on individual requests, asking twice does not speed anything up, and a request does not guarantee the page is recrawled quickly, so check the inspection result a few days later and log the date Google actually crawled it. That crawl date, not the edit date, is the start of your measurement window.
Take a snapshot before each edit as well. Save the old HTML or export the page text to the client's folder, and add the file name to the tracking sheet. When a client asks in October what exactly changed in March, you want an answer that takes ten seconds, not an afternoon in the revision history.
The tracking sheet: one tab per client
The method that makes this tutorial different from a generic refresh guide is the sheet. One workbook per client, with four tabs, kept in the client's shared folder so the account manager can read it.
| Tab | What goes in it |
|---|---|
| Selection | The 90-day export, the thresholds you used, the bucket each page landed in |
| Change log | URL, edit date, recrawl date, what changed, who checked the facts, a link to the saved old copy |
| Before / after | 28 days before and 28 days after: clicks, impressions, click-through, average position, top query |
| Controls | The same figures for the pages you left alone, over the same two windows |
Pull the figures with Search Console's date comparison filtered to one page, or export by page for each window and join them with a lookup. Annotate the edit date in your own sheet, since the report will not remember it for you. Also export the full Performance data each month. Search Console keeps 16 months of data, so a two-year comparison, or a look back at a page you refreshed eighteen months ago, depends on the exports you saved.
Add one more column if the client cares about AI search. Search Console's Generative AI performance report counts how often the site's links were shown in AI Overviews and AI Mode, and it can be filtered by page. It reports impressions only, with no clicks, so log it as a separate line on the Before / after tab rather than mixing it into the click figures. A refresh that answers the main query earlier sometimes lifts these impressions before it lifts position, and the client will want to see that.
Filled-in example for the garage client
Illustrative figures for one refreshed page, the brake pad cost guide, measured 28 days before and 28 days after the recrawl:
| Measure | Before | After | Change |
|---|---|---|---|
| Impressions | 1,420 | 1,980 | +39% |
| Clicks | 38 | 71 | +87% |
| Click-through rate | 2.7% | 3.6% | +0.9 points |
| Average position | 11.8 | 8.4 | 3.4 places better |
Now the control: four untouched pages from the same site went from 3,100 to 3,260 impressions over the same windows (+5%) and from 96 to 101 clicks (+5%). So most of the refreshed page's gain, around 34 points of impressions growth and 82 of click growth, is not explained by the site-wide drift. That comparison is what lets you tell a client "the refresh did this" instead of "rankings went up".
Keep the language measured anyway. One page is an anecdote. The claim you can make after a round of six refreshes is the average and the range: "five of six pages improved in position; the median improvement was 2.9 places; the control pages moved by under half a place." Report the page that got worse as well, with your guess at why.
Reading the result without fooling yourself
Four things distort a before-and-after comparison, and each has a check.
- Seasonality. A tyre guide will gain in autumn whatever you do. Check the same 28 days a year earlier and compare the shape, and lean on the control pages.
- Algorithm updates. If a confirmed update landed inside your window, say so in the sheet and treat that round's results as provisional.
- Averages that hide movement. Average position blends many queries. Open the page's query list and check the top three queries individually; a page can gain a new query at position 40 and make its average look worse.
- Too-early reading. Wait the full 28 days after recrawl. If the page is still moving in week four, extend to 42 and note it.
A realistic mistake: a pest-control client's page for "wasp nest removal" jumped in impressions in the first week after a refresh, and the agency reported the win. Three weeks later the control pages showed the same jump, because wasp season had started. The refresh was probably useful, but the report overstated it. The control group is what prevents that.
When to stop refreshing a page
If a page sits in the top three for its main query with a healthy click-through, a further refresh rarely pays. Move the hours to the next bucket. And if two rounds of edits leave a page still beyond position 20, the problem is probably not freshness: the page may target a query where the results are dominated by a different kind of page, such as a tool or a product listing. Consider merging the article into a stronger page, or changing its purpose. The scaled content checklist is a useful filter if you suspect the client's older pages are thin and repetitive.
Fitting a refresh round into a retainer
Agencies often underprice this work because the visible output, a page that looks similar, seems small. Scope it by the unit that matters: pages refreshed per month and a tracking report at day 28. A round for one client might look like this: week one, export and select; week two, audits and fact checks; week three, edits, client sign-off on sensitive pages and publishing; week four to eight, recrawl requests and measurement. Quote the selection and tracking as part of the service, because they are the parts a client cannot do themselves and the parts that prove value.
The client reporting tutorial covers turning the sheet into a one-page summary. Give each client the same four lines: pages refreshed, median change in position against controls, clicks gained, and what you will do next month. It stops the conversation drifting into debates about a single keyword.
A prompt pack for the editing stage
Once the audit is approved, these prompts handle the mechanical edits. Run each on one section at a time, never the whole article, and paste in the facts you have verified.
Rewrite the section below so the answer to "[main query]" comes in the first
two sentences. Keep every verified fact exactly as written. Keep the heading.
Do not add any claim that is not in the text or the facts list.
Facts verified on [date]: [list]
Section: [paste]Suggest three title tags (under 60 characters) and three meta descriptions
(120 to 155 characters) for this article. The main query is "[query]".
Each must promise only what the article delivers. Return them in a table
with the character counts.List the questions a customer of a [cleaning company / garage / furniture
maker] would still have after reading this section. Return five, ordered by
how likely they are to ask. I will answer them myself; do not answer them.The third prompt is the one agencies most often skip, and it is the one that produces genuinely new material. The answers come from the client, the writing from you, and the AI only finds the gaps.
Where the retainer can slip
Two checks keep a refresh programme honest. First, once a quarter, rerun the selection on pages you refreshed earlier. Any that have faded back into the "faded" bucket get another look, and you learn which kinds of edit last. Second, keep a running tally of hours per page against clicks gained. If a page cost four hours and added three clicks a month, the sheet will show it, and you will pick different pages next time.
The same sheet protects you in the other direction. When a client asks why nothing was refreshed on their pest-control site this month, the Selection tab shows that no page cleared the thresholds, and the quiet month is a decision rather than a lapse.
Questions agencies ask before a refresh round
Should the published date change after a refresh?
Only when the page has changed in a way a reader would call substantial. Show a visible 'last updated' date next to the original date, keep the original publication date in the page's structured data, and never bump the date after a cosmetic edit. A new date without new content misleads readers and is easy to spot.
How long should I wait before judging a refresh?
Wait at least 28 days after the page is recrawled, and compare against the same 28-day window before the edit. Seasonal topics need the same period a year earlier as a second check. Judging after a week mostly measures noise, not the edit.
Can AI rewrite the whole article during a refresh?
It can, but a full rewrite throws away what already ranks. Ask AI to propose edits section by section, keep the passages that earn impressions, and replace only what is outdated, thin or missing. A person must check every fact, price and date before it goes live.
What if rankings drop after the refresh?
Check the annotation date, then compare the pages you left alone over the same days. If the control pages dropped too, it was a wider shift. If only the refreshed page fell, restore the old headline and intro from your saved copy first, then reintroduce the other changes one at a time.
Further reads
- Best AI for SEO Content: A Refresh Workflow for Client Articles — Covers the day-to-day editing workflow for client articles.
- Search Console and AI Overviews: What Agencies Can Report to Clients — Shows what Search Console can and cannot report to clients.
- How Agencies Keep Brand Voice Consistent in AI-Written Content — Keep the client's voice intact while editing old copy.
- AI Keyword Clustering for SEO Agencies: Validate Clusters Before Sign-Off — Group queries before choosing which pages to merge.
- AI Content Workflow for Agencies: From Brief to Approved Draft — The brief-to-approval workflow a refresh round plugs into.
- SEO for AI Search: A Client Audit Checklist for Agencies — Audit the same pages for AI search visibility.
- Best AI for Content Analysis: Audit a Client's Article Library — A five-stage audit of a client's article library: export the URLs, cluster by topic, list gaps and overlaps, then rule on merge, refresh or prune.
- AI Tools and AI Development: The Complete 2026 Guide — the AI hub, including every tutorial in the AI-for-business series.
Sources: Google Search Console Help (Performance report; Generative AI performance report), Google Analytics Help (Search Console data window), Google Search Central (publication dates, asking Google to recrawl URLs, creating helpful content, analysing traffic drops).