The parts of your content that go off

Three tiers is enough. Fast: anything that changes without a decision being made about the page, which means prices, stock and availability, opening hour exceptions, staff names and roles, current promotions, and any threshold set by somebody outside your organisation.

Medium: policies, service descriptions, process explanations, delivery arrangements. These change when somebody decides they should, which means there is at least an opportunity to update the page at the same time, even if it is often missed.

Slow: explanations of how something works, background, definitions, the material that describes concepts rather than commitments. This is the safest to index and, usefully, it is also the material most people have the most of.

Do this sort on your inventory as a column. It takes twenty minutes and it tells you where all your review effort should go.

Prices are the obvious one, and the failure is rarely the headline price. It is the surrounding numbers: the delivery threshold, the deposit, the cancellation fee, the price of the add on nobody remembers is listed on page four.

Regulatory and statutory thresholds are the nastiest, because they change on somebody else's schedule and nobody sends you a note. Tax rates, allowances, minimum notice periods, eligibility criteria for public schemes, licensing requirements. If your material quotes one of these, it will be wrong at some point without anybody in your business having done anything.

Staff names are underrated. A page naming the person who handles enquiries is fine and useful until they leave, at which point it is generating messages to a mailbox nobody reads. So are stock and availability statements, which are the reason it is usually right to say what categories you carry and wrong to say what individual items you have.

The default detection mechanism is a customer complaint, and it is bad in a specific way: it only fires when somebody is affected enough to write in, which means the wrong information has already been served for however long it took to reach one persistent person.

The second mechanism is reading conversations, which is faster. A wrong answer usually leaves a visible mark in the transcript: a rephrased question, a contradiction, somebody typing that is not right. Reading conversations weekly catches most staleness within a week or two.

The third and best is a scheduled review, because it does not depend on anybody being harmed first. For the fast tier, quarterly is the outside limit and monthly is better. For the medium tier, twice a year. For the slow tier, annually, and mostly to confirm it is still relevant rather than still true.

This is the structural fix and it matters more than the review schedule. When you index a crawl of your own site, the indexed content is derived from pages you already maintain as part of running the business. When you upload a document, you have created a second copy that lives inside the tool, and that copy is now the only version of the truth that nobody will ever look at again.

The failure is completely predictable. Somebody uploads a price list at setup. Six months later prices change, the website is updated because customers see it, and the uploaded copy is not, because nobody remembers it exists. The assistant now quotes prices that contradict your own site.

So the rule is: if the material exists as a page you maintain, point at the page. Upload only what has no web equivalent, and when you do, write it in the inventory with an owner and a date so it is not invisible.

Beyond scheduling, there are two structural moves that reduce how much can go stale at all.

The first is to publish fewer perishable facts. Ranges instead of individual prices where the range is genuinely stable. Categories instead of items. We stock a range of these rather than we have four in stock. This is not evasion: it is publishing at the resolution you can actually maintain, which is more honest than publishing at a resolution you cannot.

The second is to concentrate the perishable facts in as few places as possible, for the same reason the opening hours belong on one page. If every price in your business appears exactly once, updating them is a task. If they appear in nine places, updating them is a project that will be done partially.

Some material is more dangerous indexed than absent. Anything that is superseded but still published. Internal drafts that ended up on a public URL. Old campaign pages with expired offers on them. Anything whose accuracy you cannot vouch for today.

Also be careful with anything that reads as a commitment rather than a description. Marketing copy tends to make more generous promises than policy pages do, and when it is indexed alongside the policy, the generous version is available to be quoted. Decide which one you are willing to have repeated back to you by a customer.

The practical test is short: if a customer read this passage aloud to you in a complaint, would you defend it or apologise for it. Index the first kind.