Somewhere out there, an AI assistant is recommending a product you discontinued two years ago. It’s doing it confidently, at scale, to buyers who have no reason to doubt it. And it never shows up in your reporting, because your dashboard counts mentions and nobody counts whether the mention is true.
That gap is the subject of two pieces Jasper published this week, and together they sketch a practice most marketing teams don’t have yet: the AI search audit. One post argues for a 12-week refresh cycle backed by the best dataset I’ve seen on AI citation behavior. The other asks a question almost nobody in marketing asks: is what AI says about you actually correct?
The dataset: refreshed pages win
In 2026, Seer Interactive looked at 7,683 pages cited by ChatGPT, Gemini, and Perplexity. The headline number: 75% of cited pages had been updated within the previous year. Dig one level down and it gets more interesting. Only 42% of cited pages were published in the past year. 72% were older pages that someone went back and revised.
Read that again. Recency of update, not youth of publication, is what tracks with getting cited. A page you wrote in 2023 and revised last month has better odds than one you published four weeks ago and forgot about.
If you built your career on “publish evergreen and walk away,” this stings. AI answer engines appear to weight freshness signals more heavily than traditional Google did, which turns content maintenance from an optional cleanup project into a standing cost of visibility.
The audit that follows measures three things. Visibility: are you cited. Accuracy: is it right. Freshness: when did you last touch the page.
Gauge one: visibility
Are you in the answer at all? This is the part everyone already tracks. Run a fixed set of buyer questions through ChatGPT, Gemini, and Perplexity monthly, and log which answers cite you, which cite competitors, and which cite nobody useful.
The questions matter more than the tool. Pick ten queries an actual buyer would ask, not vanity brand searches that only ever return your homepage. “Best CRM for a 20-person logistics company” tells you something. “[Your brand name] review” tells you nothing you didn’t already know.
Gauge two: accuracy
Is what the engine says true? Almost nobody has this gauge. The failure modes read like a support queue from a bad dream: discontinued products recommended as current buys, capabilities described wrong, 2023 pricing presented as today’s. Buyers have no reason to doubt a confident answer, so the error never surfaces as a complaint. It just quietly kills deals.
The fix is an audit, not an argument. Ask each engine the same standard question set about your brand. Log every wrong claim. Then trace the claim back to its source, because the source is the only place it can be fixed. The chatbot itself cannot be patched. If Gemini says your starter plan costs $29 when it costs $49, then somewhere a page, a forum thread, or a listicle says $29, and that page is the target.
One warning from Jasper that deserves emphasis: accuracy drifts. Answers are synthesized from a shifting mix of crawled content, so a correction that lands this month can silently degrade by spring. Accuracy is a queue you work, not a bug you close once.
Gauge three: freshness
When did you last touch the page? The Seer data sets the cadence: review high-priority pages every 12 weeks. High-priority means pages AI engines actually cite, plus pages where the information changes fast, like pricing, model names, and competitor comparisons. Pull pages forward in the queue when they go stale or when visibility drops.
Running it: one sheet, three tabs
You don’t need new tooling for this. A spreadsheet covers the whole loop.
- Tab one, visibility log. One row per question per engine per month. Cited, not cited, or cited a competitor.
- Tab two, accuracy log. One row per wrong claim: what the engine said, what’s actually true, the source page behind it, and a status column that stays open until the source is fixed.
- Tab three, refresh queue. Every page worth keeping, sorted by last-reviewed date, with a next-review date 12 weeks out.
Priority ordering is simple. Citation performance first, information drift second. A cited page with stale numbers is a fire. An uncited page with stale numbers is debris; clean it up when you get to it.
The opening this leaves
Most agencies still sell AI visibility as a rankings report: mentions counted, share of voice charted, invoice sent. The audit here measures three things instead of one, and two of them require ongoing work rather than a quarterly screenshot. That’s uncomfortable for vendors and genuinely valuable for clients. If you’re in-house, you don’t need an agency to start. You need one spreadsheet, ten buyer questions, and a recurring calendar block.
What to do this week
- Write down ten real buyer questions for your category.
- Run them through ChatGPT, Gemini, and Perplexity. Note every answer that cites you.
- For each citation, check the claim against reality. Log every mismatch.
- Take the worst wrong claim, trace it to its source page, and fix it there.
- Put a 12-week review of your top cited pages on the calendar. Recurring, not optional.
The first accuracy pass usually pays for itself the moment you find a discontinued product that an assistant has been happily recommending for months.
