There is a persistent myth in the AI-SEO conversation that goes something like this: AI assistants have a real-time view of the web, so your job is to publish constantly and keep dates fresh, otherwise the model will overlook you. It sounds plausible. It is also wrong in a way that costs publishers and agencies a lot of wasted effort.
This post is a clean definition of what "real-time" actually means in AI search, when freshness matters (it sometimes does), and when it is a distraction from the work that would actually move your citations.
What "real-time" really means
Almost every answer engine you care about — ChatGPT Search, Copilot, Perplexity, Google AI Overviews — works the same way at a high level:
You ask a question.
The assistant decides whether the question is time-sensitive or knowledge-grounded.
If time-sensitive, it issues one or more live queries to an external index (typically Bing for ChatGPT and Copilot, Google for AI Overviews, a hybrid for Perplexity).
It pulls the top documents the index returns.
It synthesises an answer from those documents.
The model itself does not "have" a constantly-updated copy of the web. It calls out to a search engine, reads what the search engine returns, and answers from there. "Real-time" in AI search means real-time retrieval — not real-time indexing.
The practical consequence: your AI visibility is gated by the search engine's index, not by some separate AI crawler that you must please. If Google indexed your update overnight, ChatGPT and AI Overviews can quote it tomorrow. If Google has not crawled you in two weeks, no AI assistant will surface your latest post regardless of how fresh the timestamp is.
When freshness genuinely matters
Some queries are freshness-critical. The model recognises this and weights recent documents heavily. The clearest examples:
Breaking news. "What happened with the Reserve Bank rate decision today." The assistant will explicitly look for content published in the last 24 hours.
Live events. Match scores, election results, weather warnings, traffic incidents.
Volatile data. Share prices, exchange rates, fuel prices, interest rates.
Product recalls and safety alerts. Time-bounded by definition.
Software changes. "What is new in iOS 19" — the assistant assumes you want the latest, not the historical record.
For these queries, a publish or modification date inside the last 24–72 hours is effectively a filter. Old content gets dropped from the source set, however authoritative.
When freshness is mostly a distraction
The much larger category of queries does not care about freshness — and chasing it actively wastes effort.
"What is X" explainers. "What is generative engine optimisation." The model wants depth and clarity, not a 2026 dateline.
How-to guides for stable processes. "How to claim a Google Business Profile." The process changes once every few years.
Comparison and best-of pages. Recent enough to be relevant (within a year or two), not necessarily this week.
Service pages. Your service offering does not change weekly. Pretending it does erodes trust.
Methodology and framework posts. Citation depends on depth and reasoning, not recency.
For these queries, an article that is technically two years old but genuinely thorough out-quotes a hastily-refreshed competitor every time.
The freshness matrix
A rough mental model that we use in audits:
Vertical / topic typeFreshness sensitivityNews, breaking eventsHighStocks, rates, pricesHighLocal business hours/availabilityHighSoftware release notesHighProduct comparisonsMediumLocal services (general)MediumHow-to and tutorialsLowDefinitional / glossary contentLowFramework and methodology postsLowLong-form research and analysisLow
If you publish in a "Low" row, treat freshness as a hygiene factor — keep dates accurate, update on substantive changes — not as a competitive lever. Your citations live or die on depth, structure, and authority.
What to do instead of chasing freshness
Once you stop performatively republishing evergreen content, you can redirect the effort into things that actually move citations:
E-E-A-T signals. Named author with a credible bio, organisation byline, transparent methodology, original data or examples. AI assistants strongly prefer attributed sources.
Structural extractability. Direct answer in the first 100 words, FAQ blocks, schema, semantic headings.
Depth over breadth. One canonical 2,500-word piece on a topic out-cites three 800-word fragments. Internal links pointing at the canonical page tell both Google and the AI which version to read.
Authoritative referrals. Citations from credible third-party sites do more for AI visibility than any amount of self-published refresh activity.
Article schema, used correctly
When freshness does matter, schema is how you communicate it. Two fields matter:
{
"@context": "https://schema.org",
"@type": "Article",
"headline": "South African petrol price changes — July 2026",
"datePublished": "2026-07-01T07:00:00+02:00",
"dateModified": "2026-07-03T09:30:00+02:00",
"author": { "@type": "Person", "name": "Renato Barril" },
"publisher": { "@type": "Organization", "name": "doubleBaRRiL" }
}
datePublished anchors the original article. dateModified signals that you updated it meaningfully — and for freshness-sensitive queries, this is what the retrieval layer looks at.
A few rules:
Do not touch dateModified without a real edit. Models and search engines both detect "ghost updates" (no substantive content change). It costs trust.
Show the modified date on the page, not just in schema. Visible and machine-readable dates should match.
Use ISO 8601 with timezone. South African content should publish with
+02:00. Ambiguous timestamps create extraction errors.
A small experiment you can run
If you suspect freshness is hurting you on a specific topic, run this in two weeks:
Take two comparable evergreen articles on your site. Pick topics in the "Low" or "Medium" row of the matrix above.
Pick five high-intent prompts that should surface those topics.
Query each prompt in ChatGPT Search, Perplexity, and AI Overviews. Log whether you are cited and what the assistant said about you.
Republish one article with a meaningful refresh (new examples, updated stats, expanded section), and leave the other alone.
Re-run the same prompts in a fortnight.
For most evergreen topics, you will see no movement on either article. That is the result you want — it confirms that freshness is not your bottleneck and you can redirect the time elsewhere. For the minority of topics where the refreshed article does gain ground, you have learned something specific about which of your content benefits from active maintenance.
The bottom line
"Real-time" in AI search is shorthand for "the assistant just looked it up." It is not shorthand for "the model knows what you published this morning." The retrieval layer is gated by the search-engine index, and the search-engine index already runs at the freshness cadence you need.
For freshness-sensitive content, ship accurate datePublished / dateModified, surface dates on the page, and publish quickly. For everything else — which is most of what most agencies and publishers produce — stop chasing the date stamp and invest in depth, authority, and extractability. Those are what get cited.