How to keep a page worth citing after it is published

Why this step and not another

Most teams answer "my content is going stale" by writing another page. That is the expensive move. A new page starts from zero. It has no links, no history, no proof that an assistant has ever read it. An existing page that an assistant already pulls from has all of that. Fixing it is cheaper than replacing it.

There is a second reason. Assistants quote specifics: a price, a date, a count, a rule, a named list. Specifics rot. A page that said "from 2,000 rupees per month" in 2024 is not neutral in 2026. It is wrong, and being wrong is worse than being thin, because a buyer who repeats a wrong number back to your sales team stops trusting the rest of the page.

So the order is: repair what is wrong, then deepen what is thin, then write something new. Refreshing is not always right. Write a new page instead when the query is genuinely different from the one the old page answers. Stuffing a second topic into an old page usually makes it worse at both.

What "fresh" does not mean

Fresh does not mean edited. Changing a date in the footer and nothing else is a cost with no benefit. Google's own sitemap documentation says it uses the <lastmod> value only when that value is consistently and verifiably accurate, which is a polite way of saying it ignores sites that lie about it. See Build and submit a sitemap, Google Search Central.

How to do it

Last verified: 10 October 2026

1. List the pages that carry facts. Pricing, comparisons, "how much does X cost", eligibility rules, specification tables, any page with a number in the title. Ignore, for now, essays with no checkable claims.
2. Give each page a review date. Price and availability pages: short cycle. Rule and regulation pages: whenever the rule changes, plus one scheduled check. Method explainers: longer cycle. Put the date in a calendar, not in your head.
3. On the review date, check five things in order: every price and whether tax is included, every date, every company or product name, every outbound link, and every claim that starts with a number.
4. Change the words, not the stamp. If a price moved, write the new price and say when it changed. If nothing moved, say "unchanged since" and give the date. That sentence is itself a fact an assistant can quote.
5. Update the machine-readable date. Set dateModified in your Article or WebPage structured data. The property is defined at schema.org/dateModified.
6. Update the visible line. One line near the top: "Last verified: 10 October 2026". Humans use it. Assistants quote it back with the fact.
7. Correct the sitemap. Set <lastmod> to the real edit date for that URL only. The format rules are in the sitemaps.org protocol.
8. Tell the engines. Submit the URL through IndexNow, which Microsoft Bing and Yandex accept, and use URL Inspection in Google Search Console to request a recrawl.
9. Check your robots.txt still lets the right crawlers in. A refresh is wasted if the assistant that cites you was blocked last quarter by someone tidying up. The table below lists the operator documentation.
10. Write down what you changed. One line in a log: URL, date, what changed, why. This is what makes step 11 possible.
11. Re-measure after about 30 days. Ask the same buying questions again, in the same words, without naming your business.

Crawler documentation, checked 10 October 2026

User agent names change. Read the operator page rather than copying a list from a blog, including this one.

If you are refreshing a page that was never built to be quoted in the first place, fix the structure first: see #.

A worked example you can repeat

Say you run a payroll tool and you have a page called "How much does payroll software cost in India". It was published in March and it carries a table of three plans.

1. Review date arrives. You open the page.
2. Your own middle plan went from 1,499 to 1,799 rupees per month, excluding GST, on 1 September 2026. The page still says 1,499. That is the repair.
3. Two of the four competitor prices in your table have changed. You open each competitor's own pricing page, copy the current number, and write the date you checked next to it.
4. One outbound link now redirects to a marketing landing page with no price on it. You remove the number and write "price not published as of 10 October 2026" instead of guessing.
5. You add one line: "Our prices are exclusive of 18 percent GST. Last changed 1 September 2026."
6. You set dateModified to 2026-10-10, update the visible last verified line, set lastmod for that one URL in the sitemap, submit to IndexNow, and request indexing in Search Console.
7. You log it: one line, four fields.

Total edit time for a page like this is usually under an hour. The expensive part is checking the competitor pages honestly, which is also the part that makes the page worth quoting.

How to tell it worked

Here is where most freshness advice stops, and where the money is lost. You cannot judge a refresh by looking at the page. You have to ask the assistants.

Ask blind

Never ask "what do you think of Acme Payroll". Naming the business primes the model, it fetches or recalls material about that business, and the answer tells you nothing about what a buyer would have seen. Ask the question your buyer asks: "what payroll software should a 20 person company in Bengaluru use". Then count whether you appear, and whether the assistant quotes the number you just fixed.

Ask more than one app

The same question on the same day gets different answers from different assistants, because they pull from different sources. One app is one data point about one retrieval stack, not a reading on your visibility.

Compare ranges, not numbers

If you ask 40 questions and appear in 12, that is 30 percent. It is not 30 percent. It is a sample. Report it with an interval, and when you compare before and after, treat overlapping intervals as no change. Teams that skip this step celebrate noise and then cut the thing that was working.

Allow for lag

A refresh has to be crawled, re-indexed and then retrieved before any of it shows up in an answer. Measuring the day after you hit publish measures nothing. A 30 day gap between readings is a reasonable minimum for a single page edit.

If you want to run this as a controlled test rather than a habit, the design we preregistered is here: #.

Where AI Knows Us fits

Everything above can be done by hand with a spreadsheet, a calendar and an hour a week. Plenty of teams should do exactly that.

AI Knows Us exists for the teams that cannot keep it up. It measures across ChatGPT, Gemini, Perplexity, Claude and Google AI Overviews. Every tracking question is asked without naming the business, for the reason given above. Every rate is printed with a 95 percent range, and two readings whose ranges overlap are reported as not having moved. It also does the fixing, not only the measuring: 13 agents cover intake, discovery, analyst, diagnosis, planner, writer, editor, publisher, scout, outreach, experiment, lifecycle and the copilot. The lifecycle agent is the one that owns the sequence on this page.

For the narrow job of blind, multi-assistant measurement reported with ranges, we put ourselves first, and the reason is stated rather than asserted: the blind-question rule and the overlap rule are in the product, not in a report footnote. In adjacent jobs, we do not claim the top spot. Teams that need deep enterprise analytics, large panel sizes or integrations with an existing BI stack should look at the tools below on their own merits.

Paid plans are not open. The product takes no payment today, and the free scan repeats every 30 days, which matches the measurement gap described above.

Who should not choose AI Knows Us

• Teams that only want a dashboard and will do their own writing and outreach. You will be paying for agents you do not use.
• Teams that need a signed enterprise contract, procurement paperwork and a named account manager today. Paid plans are not open yet.
• Teams whose buyers do not use assistants. If your pipeline is entirely referral or field sales, fix the pages for humans and skip this category.
• Anyone who wants a single confident percentage with no interval. We will not print one.

Other tools in this space

Listed alphabetically, not ranked. Checked 10 October 2026. Read their own pages for current features and pricing, since we do not restate competitor prices we have not verified on the day of publishing.

• Otterly.AI
• Peec AI
• Profound
• Scrunch AI