Best AI for SEO Content: A Refresh Workflow for Client Articles

For refreshing existing SEO articles, a general assistant such as ChatGPT or Claude is usually the best AI, because the work is judgement and fact-checking, not generation. Use Search Console to find decaying pages, give the assistant the old article plus current sources, then run a written pre-publish check before the update goes live.

A refresh is not a rewrite. The page already ranks for something, has links pointing at it and a URL search engines know. Most of the risk is in breaking those things or adding errors the old version did not have. So every step protects what works: pick pages on evidence, change only what is out of date or thin, keep the URL, and check the result against sources before publishing. The same steps suit a single business site as well as an agency's client list.

Follow me on Instagram@sagnikteaches

Step 1: find articles that are actually decaying

Do not refresh by gut feel or by age. Use Google Search Console, which is free, and compare a recent period with the same period a year earlier, so seasonality does not fool you.

Connect on LinkedInSagnik Bhattacharya

1. Open the Performance report, set the date range to compare the last three months with the same three months last year, and view by Pages.
2. Sort by the change in clicks. The pages with the largest drop are candidates.
3. For each candidate, open the page's queries and see which ones lost clicks or position.
4. Check whether impressions fell (demand or ranking dropped) or only clicks (a competitor or an AI answer takes the click). Search Console's Search Generative AI performance report, rolled out to all sites by 31 August 2026, shows impressions from features such as AI Overviews and AI Mode but no clicks, so it will not tell you how many clicks AI answers took.
5. Skip pages that fell because the topic is gone, such as a retired product. Those need a different decision: merge, redirect or remove.

Export the list to a spreadsheet with columns for URL, clicks then and now, change, top lost query and your note. That spreadsheet is the refresh queue, and it is the first thing to show a client who asks why you are touching old pages.

Subscribe on YouTube@sagnikteaches

Step 2: decide what kind of refresh each page needs

Not every decaying page needs the same fix, and the cost of overdoing it is rewriting something that was fine. Sort pages into four types.

Only the first three are real refreshes. The fourth changes URLs and needs a redirect plan, so do it deliberately and one pair at a time. A page that fits none of these may simply need an internal link or two, which costs nothing.

Step 3: build a brief from the old article, not from scratch

Give the assistant the existing article and ask it to audit rather than to write. The audit prompt is the most valuable part of this workflow, because it produces a change list you can review before any new text exists.

You are an editor refreshing an existing article for a client. Do not rewrite it yet.

1. List every factual claim that could be out of date (prices, product names, features, dates, statistics, laws, links). For each, give the sentence, why it may be stale, and what I should check and where.
2. List the headings and say what a reader searching for "[target query]" would expect that is missing.
3. List sentences that are vague, repetitive or generic.
4. Propose a change list, ordered by impact, and mark each change as Fact fix / Add section / Tighten / Title or meta.
5. List anything in the article that must NOT change (the URL, the main heading's intent, the sections that already answer the query well).

ARTICLE:
[paste, with the URL and the target query]

The "must not change" item is what keeps the refresh from breaking a ranking page. Ask for it every time. And note the instruction not to rewrite yet: if you let the assistant produce a full new version straight away, you lose the chance to decide what changes.

Worked example: a client article about project software

Illustrative. An agency maintains the website content for a small B2B software reseller. One article, "Best project management tools for small teams", was published two years ago and has lost 38% of its clicks year on year. Search Console shows the lost queries are "project management software pricing" and "free project management tools".

The audit prompt returns, in outline: seven prices and plan names flagged as likely out of date; two tools mentioned that have since been renamed; a missing section on AI features, which competing pages cover; a vague opening paragraph; and a note that the heading "Best project management tools for small teams" and the URL must not change. The assistant proposes eleven changes and marks four as high impact: fix the seven prices, rename the two tools, add a short "AI features" section and rewrite the opening.

The editor then does the checking the assistant cannot: opens each vendor's pricing page, copies the current price and date into the spreadsheet, and tells the assistant to update only those sentences with those numbers. Time: about 25 minutes for the audit and prompts, and 75 minutes for the vendor checks, against perhaps four hours to rewrite from scratch. The refreshed article is published at the same URL with a new modified date and a note in the client's change log.

Four weeks later the agency compares clicks for the two lost queries with the weeks before the refresh. If the page has recovered, the log records the refresh as a success, and the method goes in the playbook. If it has not, the next question is whether the cause was content or something else, such as a stronger page elsewhere. Resist refreshing the same page twice in a month; give search engines time to re-crawl and settle.

Step 4: update with AI, one change at a time

Work through the change list in chunks, not in one giant prompt. A chunk is a section or a paragraph. For each, give the assistant the old text, the verified facts and the instruction.

Update the section below. Use ONLY the verified facts I give you. Keep my headings, voice and sentence rhythm. Change as little as possible. Do not add statistics, quotes or claims I have not supplied. If a sentence cannot be updated from my facts, leave it unchanged and flag it with [CHECK].

VERIFIED FACTS (with source and date):
- Tool A: Starter plan $X a month billed annually (vendor pricing page, [date])
- Tool B: renamed to [new name] on [date] (vendor announcement)

SECTION:
[paste]

A sample output (illustrative): the paragraph returns with two prices changed and the old tool name replaced, plus a "[CHECK]" flag on a sentence claiming "most teams prefer Tool B", which the assistant could not support. That flag is the useful part: it points at a claim that was never sourced. Decide whether to find evidence or delete the sentence.

Adding a new section is a different job. For the AI-features section, give the assistant two or three vendor pages as sources and ask it to summarise only what those pages say, with the page named after each claim. The same method is used in ChatGPT prompts for SEO content writing, and for the brief side, a quality gate for clusters, briefs and meta descriptions shows what a good brief contains.

Step 5: the pre-publish check

This is the step that separates a refresh from a content mill, and it is where Google's guidance points: generative AI can help with research and structure, but output can be inaccurate, so all of it should be reviewed and fact-checked, including titles, meta descriptions and structured data. Make the check a written checklist that a second person signs, and attach it to the ticket.

The diff is the key artefact. Ask the assistant to produce a before-and-after list of every sentence changed, or use your CMS's revision comparison, and review that rather than the whole page. A reviewer who reads only changed sentences spots invented claims far faster than one who rereads the whole article.

How an AI refresh quietly breaks a ranking page

• Silent rewording of correct facts. The assistant "improves" a sentence and changes a number. It shows up in the diff, and only if you read the diff. Fix: instruct "change as little as possible" and review changed sentences.
• Invented support. The refreshed text adds "according to industry research" with nothing behind it. It shows up in a search for phrases such as "research", "studies" and "most". Fix: delete or source them.
• Lost specifics. A generic rewrite removes the distinctive example that made the page rank. It shows up when you compare the old and new sections. Fix: list sections that must not change in the audit.
• Tone drift. The updated paragraphs sound like a different writer. Fix: paste a sample paragraph the client likes and ask the assistant to match it.
• Refreshing for freshness only. Changing the date without changing the substance. Do not update the modified date unless the content genuinely changed; it can mislead readers and gains nothing.

Scale makes these errors worse. If you refresh fifty client pages a month with a thin check, you are running the pattern Google's spam policy describes as scaled content abuse. An AI content checklist for SEO agencies goes through how to stay on the right side of that line.

Prioritising the queue: a simple score

A list of thirty decaying pages is more than a week's work, so rank it. A workable score multiplies three things, each rated 1 to 3: how much traffic the page lost, how close the page still is to a good position, and how commercially useful the page is to the client. Illustrative: a page that lost 400 clicks a quarter (3), still ranks around position 8 (3, close enough to recover) and attracts buyers (3) scores 27. A page that lost 20 clicks (1), sits at position 40 (1) and attracts casual readers (1) scores 1. Start at the top, and leave the bottom for a quarter when nothing else is pressing.

Ask the assistant to compute the scores from your spreadsheet rows if you wish, but supply the ratings yourself, since they are judgements about the client. And check the sort order by eye. A column of numbers sorted wrongly is a surprisingly common AI error.

A second example: a statistics-heavy article

Illustrative. A client's article on "how much does a commercial roof repair cost" quotes four figures from a 2021 industry page. The audit prompt flags all four as likely stale, and notes the article states each as a flat fact. The editor cannot find current equivalents from a source that publishes methodology. The honest options are to replace the figures with ranges from the client's own recent quotes, to attribute them as "recent quotes from our team" if the client agrees, or to remove them and explain how cost is built up from area, access and materials.

The editor chooses the third and sends the client a short note: "We could not verify the 2021 figures against a current source, so we removed them and added a section explaining what drives the price, with a prompt for readers to request a quote." The assistant drafts the new section from the client's own description of how it prices jobs. The page loses a statistic and gains accuracy. This is the right trade: an unsupported number is a liability, and a page that explains cost honestly often earns the click anyway.

Title and meta refreshes: the cheapest wins

When impressions hold but clicks fall, the cheapest change is often the title and meta description. Ask the assistant for five variants and choose by hand.

Here is the current title and meta description for an article, the target query, and the three queries that lost clicks. Write five title options under 60 characters and three meta descriptions of 120 to 155 characters. Each must be accurate to the article's content, include the main query naturally, and avoid hype, clickbait and unverified claims such as "best" or "number one" unless the article supports them.

A sample result (illustrative): the old title "Best Project Management Tools" becomes "Project Management Tools for Small Teams: Plans and Prices Compared", which states what the page contains. Check the new title against the article, because a title that promises "Prices Compared" on a page with no prices is a quick way to lose readers. Change one page's title at a time where you can, so you can see what moved, and note the date.

What this costs in time for an agency

Illustrative figures for a mid-sized article. Audit prompt and change list: 20 minutes. Fact checking at source: 45 to 90 minutes depending on how many vendors are involved. Section updates with the assistant: 30 minutes. Pre-publish check and diff review: 30 minutes. Publishing and logging: 15 minutes. That is two to three hours per article, against four to six for a rewrite, and it produces a smaller, safer change. The fact checking is the largest block and cannot be skipped; if it takes less than half an hour for an article full of prices, you are probably not checking at source.

Keeping the URL, and what to do when you cannot

The default is simple: the URL does not change, the canonical tag stays pointing to it, and internal links keep working. You change the content, the title and meta if needed, the modified date, and add internal links from newer pages. That preserves the history a ranking page has built.

There are three cases where the URL question arises. If the page's slug contains a year ("best-tools-2023"), resist the urge to change it; update the title and keep the slug, or accept the dated slug. If two pages should merge, choose the stronger URL as the survivor, redirect the other to it permanently, and update every internal link. And if a page is truly obsolete, redirect it to the closest relevant page, not the home page. In every case, log the change and re-check the redirect after publishing.

Choosing a tool: when a general assistant is enough

A general assistant covers the audit, the section updates and the meta drafts. A plan costs little: ChatGPT Plus is $20 a month, and Claude Pro is $20 a month or $17 billed annually, on Anthropic's pricing page in November 2026. For a team, ChatGPT Business and Claude Team both start at two seats, and both list a standard seat at $20 a month billed annually or $25 monthly.

SEO platforms add what an assistant lacks: keyword volumes, competitor page data and rank tracking. Semrush's pricing page in November 2026 listed SEO at $139 a month ($117.33 billed annually), Starter at $199 ($165.17 annually) and Pro+ at $299 ($248.17 annually), with Advanced and Enterprise on request. Content-optimisation tools add on-page scoring, such as the Content Score in Surfer's Content Editor. Whether they are worth the money depends on volume. A rule of thumb: if you refresh fewer than ten pages a month, the assistant plus Search Console is enough; if you manage many clients and need competitor data per page, an SEO platform earns its place. Plans change, so test any tool on three of your own articles and measure whether it found changes you would not have found by yourself.

For a single-page walk-through of on-page improvements, see AI content optimisation for SEO, one page step by step, and for deciding which pages deserve attention first, auditing a website's SEO with AI in one afternoon.

Reporting the result to the client

A refresh programme is easy to doubt, because results take weeks. Report in a fixed format so the client sees process and outcome.

1. Before: the page, the clicks and impressions for the comparison period, the lost queries.
2. What changed: the diff summary in five bullets, with each fact verified and dated.
3. Date published: and the date you will review.
4. After: the same metrics four to six weeks later, with a plain note if nothing has changed yet.
5. Next action: keep, refresh again, or merge.

Be honest when a page does not recover. A refresh is a bet on a cause; when the bet is wrong, the log tells you what to try next. If you would like to build this process for your own team, that is what my AI implementation consultation covers, from the audit prompt to the sign-off checklist.

Questions agencies ask about AI content refreshes

Which AI tool is best for refreshing existing articles?

A general assistant such as ChatGPT or Claude is enough for most refreshes, because the hard part is deciding what to change and checking facts, not generating words. SEO platforms add keyword and competitor data. Choose by whether you need that data, and test on three articles first.

Should I change the URL when I refresh an article?

No, not unless there is a strong reason. A ranking page keeps its URL, and you update the content, the modified date and internal links in place. Changing the address risks losing the rankings and links the page has earned. If a URL must change, set up a permanent redirect.

Will Google penalise AI-assisted updates?

Google's published guidance says generative AI can help with research and structure, but using it to produce pages that add no value may breach its spam policy, and all output should be reviewed and fact-checked. The test is whether the refreshed page is accurate and useful, not which tool touched it.

How often should client articles be refreshed?

Refresh on evidence, not on a calendar. Check Search Console quarterly for pages whose clicks or impressions have dropped against the same period a year earlier. Fast-moving topics, such as prices and software, may need checking more often than evergreen ones.

Further reads

• How to Choose an SEO Agency That Understands AI Search — What clients should ask an agency about AI search.
• How to Write Website Articles With AI That Still Rank on Google — Writing new articles that still rank.
• What AI Can and Can't Do for Your Local SEO — Where AI helps and stops for local SEO.
• Best AI for Writing Copy: A Freelance Copywriter's Edit Workflow — An edit workflow for AI-assisted copy.
• Best AI for Content Analysis: Audit a Client's Article Library — Auditing a client's whole article library.
• AI Real Estate SEO: Neighbourhood Pages That Avoid Steering — How agents can draft neighbourhood pages with AI without steering language: verified facts, a page template, wording rewrites and a review routine.
• Content Refresh With AI: Update Old Articles and Track Rankings — A client-by-client method for choosing old articles, refreshing them with AI, and measuring the ranking change against pages you left alone.
• AI Tools and AI Development: The Complete 2026 Guide — the AI hub, including every tutorial in the AI-for-business series.

Sources: Google Search Central guidance on generative AI content and its June 2026 announcement of the generative AI performance reports; Semrush, Claude, ChatGPT and ChatGPT Business pricing pages; Surfer Content Editor page; all checked November 2026.