How to scale content refresh

This is the agency version of scale content refresh: fewer theory, more sequence. It assumes you have a site, a product page and at least one person who can publish. If you are starting from zero, do the beginner steps first; the advanced steps will not work without a baseline.

Before you start

• A list of 50 prompts in your buyers' words, with the wording frozen.
• Access to the CMS and the schema layer (or the built-in generator).
• One owner and two hours a week for eight weeks.
• A baseline you are willing to be judged against.

The agency walkthrough

1. 1
   Build the prompt set

   Pull questions from sales calls, support tickets and the engines themselves. Group them by intent: learn, compare, buy, brand. Fifty is enough to start.

2. 2
   Baseline every engine

   Run the set in ChatGPT, Google AI Overviews, Perplexity and the rest. Record the answer, the cited URL and the claim. Export it; the baseline is the only defence against revisionism.

3. 3
   Pick the ten gaps that matter

   Rank by revenue proximity, not by volume. A branded prompt that gets the price wrong outranks a generic prompt with no buyer behind it.

4. 4
   Fix the page that should win

   Add the missing section, the named entity, the proof point and the date. If no page exists, write one page, not ten.

5. 5
   Add the machine layer

   Organization, Product and FAQ schema where they fit. An llms.txt that points at the pages you want cited. A robots.txt that does not block the crawler.

6. 6
   Publish and re-check

   Re-run the same prompts after 14 days. Log what changed. If nothing changed, check crawl access before rewriting again.

7. 7
   Widen only after stability

   Add 25 adjacent prompts once the first 50 show a trend. Growing the set before the baseline is stable hides the signal.

The checklist

How to measure it

Common mistakes

• Rewriting a page without capturing the baseline first.
• Adding schema that describes a page you have not written yet.
• Blocking GPTBot in robots.txt and then wondering why citations do not change.
• Judging a four-week programme on a 48-hour window.

Frequently asked questions

Methodology and sources

• Citekit product and pricing data, 9 November 2026.
• Vendor pricing pages and product documentation, checked November 2026. Verify current pricing with each vendor before purchase.
• Public documentation for schema.org, llms.txt and the AI engines named on this page.

How Citekit compares

Citekit is measured against Profound, Scrunch AI, Otterly.AI and nine other tools in the category. The table is the short version; every comparison has a full, sourced page.

Full comparisons: Profound · Scrunch AI · Otterly.AI · all 12 competitors

Keep reading