Answer engine optimisation for a public body starts with what its own site leaves open to machines. Google says a page must be "indexed and eligible to be shown in Google Search with a snippet" to appear as a supporting link in AI Overviews or AI Mode, so a superseded page that stays indexed stays in the pool.
The Department for Business and Trade (DBT) described a case in April 2026: an AI overview gave a wrong fee in answer to a question about the cost of setting up a charity, and its authors say "the GenAI bot pulled information from an unmaintained, unused GOV.UK page that mentioned 'charity'". The causes we can see as a builder and host sit in templates, defaults, metadata, redirects and the publishing workflow, and so do the fixes.
Why do AI Overviews quote pages an organisation has already replaced?
AI Overviews can quote a replaced page for as long as that page remains published and indexed. Google's guidance on AI features sets indexing and snippet eligibility as the conditions for a supporting link, and sets no date or freshness condition.
DBT's authors explain the exposure: "GenAI bots are designed to search and gather information from all corners of the internet". Their response was a manual audit of GOV.UK pages, including pages published by the former Department for Business, Energy and Industrial Strategy and never transferred to DBT. A page qualified when it had "not been updated in 5 years", had "fewer than 11 views in 5 years", was a content type meant to carry current information, and had "no owner". 150 pages met all four criteria, and DBT redirected them to the National Archives copy, a GOV.UK page that better meets the user need, or the new legislation the old guidance was based on.
The Register reported the work on 23 April 2026 as stale gov.uk pages feeding AI overviews old data. That generalisation is The Register's; the audit and its example are DBT's.
The ownership criterion is the one a platform cannot fix on its own. A separate DBT post on content ownership puts it plainly: "When ownership is unclear, the default can become to keep everything, not through intent but through uncertainty." Devolved publishing spreads ownership across more teams, as covered in the hidden cost of decentralised content publishing.
Does a withdrawn banner keep an old page out of AI answers?
A withdrawn banner does not keep an old page out of Google's AI answers, because the page stays indexed and in Google's results. GOV.UK's guidance on retiring content says "Withdrawing content means it's still available at the same URL", and that withdrawn content will still appear in "Google search results, and attachments can still be reached directly from Google".
DBT considered banners and chose redirects. Its user research participants would "just skip over the banner", and the team wanted to make sure GenAI bots would not pick the pages up. GOV.UK's unpublish option takes a page off the site completely and requires an alternative URL to redirect users to.
We use a banner to tell readers a page is out of date, and a redirect, a noindex rule or removal to keep it out of search indexes. The table is our working rule for council and agency sites. It is not Google's or GDS's.
| Page status | What the reader needs | Platform control |
|---|---|---|
| Draft or pre-publication | Nothing yet | Behind a login |
| Superseded, with a replacement | The replacement | 301 redirect |
| Withdrawn, kept for the record | Context | Banner plus noindex |
| Time-limited, with a known end date | The current offer | unavailable_after |
| Current | Proof it is current | Visible dates matching structured data |
Choosing between a redirect and a 410 for a page with no replacement is covered in website decommissioning.
How do we keep draft pages out of ChatGPT, Perplexity and AI Overviews?
Draft pages stay out of AI answers when they sit behind a login. Google's guidance on controlling what you share says password protection will "prevent that content from appearing in Google Search", while its robots.txt guide says robots.txt "is not a mechanism for keeping a web page out of Google."
The AI vendors' own documentation differs on robots.txt. OpenAI says its ChatGPT-User agent acts on user requests, so "robots.txt rules may not apply." Perplexity says Perplexity-User "generally ignores robots.txt rules" because a user requested the fetch. Anthropic states that its bots honour robots.txt directives. Robots.txt also blocks the crawler from seeing a noindex rule: Google says a page carrying noindex must not be blocked by robots.txt, or it can still appear in results.
On the platform side, a draft leaks when a CMS issues preview URLs that work without signing in, or when a staging copy runs on a public hostname. Crawler traffic has a cost of its own, set out in what bot traffic costs you.
Consultations stay published, so the page has to state its stage
GOV.UK's retirement guidance says "Do not withdraw consultations unless you know that a subsequent consultation directly supersedes an existing one." Under that guidance a closed GOV.UK consultation stays live, with proposals that may never have become the rule. We found no Google guidance specific to consultations, so what follows is our recommendation using its general controls.
Have the template print the status and closing date near the top from a status field, and link the outcome there once it exists. Google's robots meta tag documentation offers data-nosnippet for "textual parts of an HTML page not to be used as a snippet", which lets a template keep the proposal text out of snippets while the page stays findable. The same documentation says the page-level nosnippet rule "will also prevent the content from being used as a direct input for AI Overviews and AI Mode." It states no AI effect for data-nosnippet, which Google's AI features page lists among the controls that limit what is shown from a page in Search.
Which dates should a page show for answer engine optimisation?
A page should show a visible, labelled published or last-updated date, and carry the same dates as datePublished and dateModified in Article structured data. That is Google's byline date guidance, which also asks that the structured dates match the visible ones.
Its guidance on helpful content asks: "Are you changing the date of pages to make them seem fresh when the content has not substantially changed?" For sitemaps, Google uses <lastmod> only when it is "consistently and verifiably" accurate, and a copyright date change does not count.
A CMS that rewrites dateModified on every save, or stamps a new date when a footer block changes, shows a new date when the content has not substantially changed, and a sitemap that takes <lastmod> from the same field is no longer verifiably accurate. Fixing that is a template change. The sitemap should list only the preferred URL for each current page, which matters on Drupal sites where taxonomy term pages can publish duplicate routes to the same content.
Should we switch off AI Overviews with Google's new Search Console control?
Google's Search generative AI control applies to a whole Search Console property, so turning it off removes current, correct pages from AI Overviews and AI Mode along with the stale ones. Google says that as of 31 August 2026 it has rolled the control out to all websites worldwide.
The control covers AI Overviews, AI Mode and generative AI features in Google Discover, and child properties can override it. Google says it "isn't used as a ranking or inclusion signal affecting other parts of Search". AI training sits outside it, under Google-Extended, which Google says does not affect a site's inclusion in Google Search.
In our view a property-wide switch suits a section that should never feed AI answers. For stale pages, the page-level controls in the table do the job without losing the pages that are right.
What to ask a supplier about answer engine optimisation
- How do drafts, previews and staging stay off the open web? A strong answer names authentication on each. A weak one says "we block them in robots.txt".
- What does the template do when content is withdrawn or superseded? A strong answer describes a status field that drives the banner, the robots meta tag and the sitemap together. "Editors can add a banner" is weak.
- When does
dateModifiedchange? A strong answer ties it to a change in main content and shows it matching the visible date. "On save" is the weak answer. - Which URLs does the XML sitemap list? Look for current, canonical pages only, with withdrawn and noindexed pages excluded automatically.
- How are redirects kept and audited after content is retired? A strong answer names where the redirect map lives and who reviews it.
How Axistwelve builds for this
We build status, dates and retirement rules into Drupal templates and content models, so the platform applies them rather than the editor. For organisations that run AI answers on their own site, Axistwelve Semantic Search is built to keep withdrawn, draft and expired pages out of those answers.
Before you run an answer engine optimisation check
- Ask five AI tools, such as ChatGPT, Perplexity, Gemini, Claude and Google AI Mode, five questions your organisation is the authority on, and compare each answer with your current page.
- Trace each wrong answer to the page it came from, and decide between redirect, noindex, update or removal.
- Check a sample of drafts, previews and staging URLs from a browser that is not signed in.
- Name an owning team for every content type that carries current rules or fees. If AI tools also answer questions on your own services, read what councils need in place before using AI for citizen services.
Frequently asked questions
Does blocking GPTBot keep our pages out of ChatGPT search?
Blocking GPTBot opts a site out of OpenAI model training, not ChatGPT search. OpenAI's crawler documentation says OAI-SearchBot is the agent "used to surface websites in search results in ChatGPT's search features", and each setting is independent.
Do public sector sites need special markup or an AI text file to appear in AI Overviews?
Public sector sites need no special markup or AI text file to appear in AI Overviews. Google's AI features guidance says "You don't need to create new machine readable files, AI text files, or markup to appear in these features."
How long does it take for a corrected page to show in AI answers?
Google's AI features page gives no timescale for a corrected page to reach AI answers. On changes to preview controls, it says "crawling can take anywhere from several days to several months". A redirect or noindex on an old page also waits on a recrawl.
