Give your AI workflow a date with a defined meaning
Capture a reference timestamp once, name the publishing timezone and keep event dates separate. Each field needs an explicit role in the article.

A workflow can give an AI model the current timestamp. The harder part is making every date in the output mean the same thing as that timestamp.
A news summary may need the reader's local date. An execution log usually needs an unambiguous instant. An article about a historical event needs the event's date even when it is written today. Treating all of those as one field creates errors that a stronger prompt cannot repair.
Capture a reference time explicitly
n8n provides the current time through its built-in Luxon object, $now. Its date-handling documentation also describes $today, which is rounded to the start of the day. For a reference instant, use the full timestamp.
In a field that supports n8n expressions, this produces an ISO timestamp in UTC:
{{ $now.setZone('UTC').toISO() }}
Evaluate it in an early Edit Fields node and store the result in a field named referenceTimeUtc. Pass that field to later nodes as part of the execution data. Capturing it once gives a long execution a consistent reference even if later steps run after midnight.
Do a manual run to inspect the evaluated value. The output should contain an actual timestamp, with an explicit UTC offset, before it reaches the model. If the model receives the expression's braces as plain text, the expression has not been evaluated in the intended field.
Give the local date a named timezone
The same instant can belong to different calendar dates in different places. Choose the timezone for the publishing task separately from the storage format of its execution record.
For a workflow serving a reader in Istanbul, Europe/Istanbul is the relevant timezone name. Use it when converting the captured instant into a display date. Keep the original UTC instant so a later review can reproduce the conversion.
n8n's workflow settings include a workflow-specific timezone. Check that setting when a scheduled publication fires on an unexpected date. Changing the wording in the AI prompt leaves the trigger's configuration untouched.
For repeatable output, pass the chosen timezone beside the reference time. A field called today without a timezone is ambiguous to someone inspecting the record later. A named zone makes the rule explicit, with no need to infer it from the server's location.
Keep event dates independent
Consider an illustrative article written on 14 November 2026 about a product released in December 2025. The writing date is in 2026. The release remains a 2025 event. A source's last-modified date may describe a subsequent documentation edit, with no change to when the release happened.
Store those meanings in separate fields. The reference timestamp records when the workflow anchored its task. An event date describes what happened. A source consultation date records when you checked a page. Replacing every old year with the current year corrupts the article.
Relative phrases need the same discipline. A quotation that says next year inherits the speaker's date. Paraphrasing it with the intended calendar year can remove ambiguity, provided the source establishes that year. If the source's timing is unclear, keep the uncertainty in the text.
A model can still generate an incorrect temporal claim after receiving a correct reference time. Review the sentence's meaning against its source. A factual claim that a release occurred still needs evidence from its source.
Format before passing dates onward
n8n passes dates between nodes as strings. Its documentation recommends parsing those strings when date arithmetic is needed. That is especially relevant when a later node compares an expiry date with the captured reference time.
The date-expression reference documents ISO output and date operations. Use the explicit conversion for the format your input actually contains. A day-month-year string cannot safely be interpreted through a guess about the reader's locale.
Keep a timestamp with its offset for machine comparisons. Derive a shorter display date only at the point where the presentation needs it. Discarding the offset early makes later diagnosis harder because the original instant has been lost.
The host's clock is another input to inspect. A timezone conversion changes how an instant is displayed. It cannot repair a system clock that reports the wrong instant. If workflow records disagree with an independent clock, investigate the host before changing the article prompt.
Review temporal claims before publication
A useful test set includes an execution close to local midnight and another close to the end of a year. Check what the model receives, then read every relative date in its output. The expected result should follow the task's chosen timezone.
Automated checks can flag suspicious years for review. Their outcome should be a flagged sentence with its context. A historical article can legitimately mention many previous years, so deleting old dates through a blanket rule would destroy useful information.
For publishing, retain the reference timestamp with the reviewed source version. That lets you reconstruct why the workflow described an event as past or forthcoming. Capture time explicitly, keep its meaning attached to the field, then verify the factual claim that uses it.