Lawyers ask two versions of this question: "why does ChatGPT still show our old address?" and "how fast will our new content show up?" Both answers depend on which of ChatGPT's three information layers produced the response — the trained model, the live search index, or a specific fetched page. Each updates on its own schedule, and each has its own fix.
Layer 1: the trained model — slowest, months to years
The base model's knowledge is frozen at its training cutoff. When ChatGPT answers without searching — which still happens for general questions — it draws on what the web said about your firm as of that cutoff, which may be a year or more old. You cannot edit this layer directly. What you can do is make sure the web's persistent record of your firm (your site, directories, Google Business Profile, professional profiles) is accurate and consistent, because that record is what future training runs absorb. Firms that cleaned up their entity footprint in one year typically saw the corrected version reflected in the following generation of models.
Layer 2: the search index — days to weeks
When ChatGPT searches — automatic for anything it deems current, local, or specific — it queries a live index crawled by OAI-SearchBot. This layer behaves like a young search engine: new pages from established sites typically appear within days to a couple of weeks, and updates to existing pages are picked up on recrawl. Sitemap hygiene, clean canonicals, and internal links from frequently-crawled pages all shorten the cycle, exactly as they do for Google. If your new practice-area page is not surfacing after a month, verify OAI-SearchBot can access it and that it is in your sitemap.
A note on directories: the layer you forgot you had
Between your website and the model sits a stratum of third-party pages — legal directories, chamber listings, review platforms, old news mentions — that both the search index and past training runs have read. These pages age badly because nobody owns updating them. The associate who left in 2022 still headlines a directory profile; the office you moved out of still anchors a citation page; a practice area you exited still leads a listing description. When ChatGPT asserts something wrong about your firm without citing a source, one of these fossils is the usual culprit, because the model saw it repeated across several aggregators and treated repetition as confirmation.
The remedy is an annual sweep: search your firm name plus each old address, former lawyer, and discontinued service, claim or correct every listing you find, and request removal where correction is impossible. It is tedious once and cheap forever — and it cleans the record for every engine at the same time, not just ChatGPT.
Layer 3: the fetched page — real time
When an answer cites your site, ChatGPT often fetched that page during the conversation. Whatever the page said at that moment is what the user was told. This is the layer firms underrate: your website is being read aloud, live, to prospective clients. A stale fee, an old address, a discontinued service on a live page goes straight into answers the same day — and so does a correction. Keep the pages engines cite most rigorously current; they are your real-time spokespeople.
What speeds the cycle up — and what quietly slows it down
Recrawl frequency is earned, not requested. Pages that change on a predictable rhythm get visited more often; pages that sat unchanged for three years get visited on a three-year assumption. A firm that updates its core practice-area pages quarterly — refreshed figures, a new dated FAQ, a revised example — trains every crawler, OAI-SearchBot included, to come back sooner. Internal links matter the same way they always have: a new article linked from your most-crawled pages is discovered in days, while an orphaned page linked from nowhere can sit unindexed for weeks.
The quiet decelerators are technical. Redirect chains from old URL migrations, canonical tags pointing at the wrong variant, and sitemaps still listing deleted pages all waste the crawl attention you have. Slow server responses do too — crawlers budget their time per site, and a site that serves pages in four seconds gets fewer pages read per visit than one that serves them in four hundred milliseconds. None of this is AI-specific advice, which is the point: the search layer of ChatGPT inherits the crawl mechanics of the web, so classic technical hygiene is now also freshness strategy for AI answers.
Correcting wrong information, layer by layer
- Wrong in a cited answer: fix the cited page (yours) or request a correction (third-party page), and the next fetch reflects it
- Wrong from the search layer with no citation: update your site and high-authority profiles, then allow a recrawl cycle
- Wrong from the trained model: fix the persistent web record everywhere it appears, and be patient — this layer only moves with new model releases
- Wrong everywhere: your entity footprint disagrees with itself; run a consistency audit before anything else
The practical monthly habit: ask ChatGPT your firm's name plus your top three services, note anything wrong, and trace it to its layer. Most "ChatGPT is wrong about us" complaints turn out to be a third-party directory page the firm forgot existed — findable, and fixable, in an afternoon. Assign the check to whoever owns your intake reporting, put it on the same monthly calendar, and wrong answers stop surviving long enough to cost you a client.
Frequently Asked Questions
Grow your ChatGPT practice with AI
Lexscale.ai builds AI search visibility, websites, and intake systems for chatgpt firms across North America. Book a free strategy call to see what would move the needle for your practice.
Book a Free Strategy Call →