Common AI mistakes with dates and times, and how to spot them
Language models often struggle with days of the week and relative dates. Learn about their most common mistakes and how to spot them in good time.
The short answer: language models do not check an internal clock or calculate calendars; they predict text word by word. That is why they often misinterpret relative expressions like 'next Thursday', mismatch days of the week with calendar dates, and overlook time zone differences. The best way to catch this is to always check the original message's date, verify that the day and date match in your calendar, and avoid letting AI schedule appointments autonomously.
Delegating message triage to extract commitments saves time, but it requires knowing where the technology stumbles. If you understand how a smart assistant struggles with processing time, it becomes far easier to catch inaccuracies before you miss a meeting.
The lost time anchor: older messages with relative terms
The most common issue arises when a message uses relative expressions ('tomorrow', 'this Friday', 'next week'). If a client emailed you on Tuesday saying 'I will send you the draft tomorrow', the intended day was Wednesday.
However, when AI analyses that email days later—for instance, during a morning summary on Thursday or Friday—it can make two mistakes:
- Taking the moment you read the summary as its reference point, assuming that 'tomorrow' means the day after today.
- Assuming the message date is irrelevant and omitting the actual deadline because it presumes the task is already overdue without checking.
To spot this, always look at the header of the original email. If a summary mentions a relative deadline, check the date the note was originally sent. As we explained when discussing what an AI secretary actually does, human judgement to verify context remains indispensable.
The day-of-the-week hallucination
Artificial intelligence models struggle significantly with modular arithmetic, the mathematical foundation used to work out which day of the week corresponds to any given date. It is remarkably common for an AI to produce entirely plausible yet false statements, such as 'the meeting on Tuesday, 17 November', when the 17th is actually a Wednesday.
This error usually stems from one of two causes:
- A typo in the original text: if the sender mistyped the day or the date, AI rarely corrects it; it tends to pick one at random or perpetuate the inconsistency.
- Pure model hallucination: when summarising a message that only mentioned '17 November', the system helpfully adds the day of the week to sound more natural, miscalculating in the process.
To catch this mistake, get into the habit of checking the date on your calendar rather than relying on the day of the week suggested by the generated text.
Confusion over the ambiguity of 'next'
In everyday language, saying 'next Friday' on a Wednesday can mean two different things depending on the speaker or region: the Friday of that same week (in two days) or the Friday of the following week (in nine days).
AI has no way of discerning your correspondent's linguistic habits. It usually defaults to whatever pattern is most common in its training data, leading it to schedule appointments a week early or a week late. Moving from an email to a dated commitment requires you to check the thread's context or ask for explicit confirmation with a numerical date whenever there is any doubt.
Time zones and international formats
Another classic blind spot is unspecified time zones. If an email mentions a 'call at 10:00' and the sender works in London or New York, the AI often assumes the stated time refers to your local time zone, or it might convert the time twice if the email client had already adjusted the timestamp.
Similarly, when numerical date formats are mixed (such as the American month/day/year format versus the British and European day/month/year convention), '04/05' can easily be read as either 4 May or 5 April. If you work with international teams, always look for the UTC offset or the full name of the month.
How to safeguard your daily routine
To stop these slips from disrupting your day, apply these quick checks:
- Review before scheduling: use AI to draft replies or identify action points, but make sure you confirm adding the event to your calendar yourself.
- Check your calendar visually: a quick glance at a unified view of your day's tasks and appointments will immediately show if a phantom meeting has clashed with your plans.
- Decide which tools handle your data: with Madrugo, for example, the morning briefing brings together your email, calendar, and tasks using AI, but if you prefer not to rely on these models, AI can be switched off completely so the summary is generated on the server using deterministic rules. What is more, the privacy policy available at our privacy policy sets out how and where data is processed so you remain in control at all times.
Frequently asked questions
- Why does artificial intelligence mix up days of the week and dates?
- Because language models do not carry out real-time calendar calculations; they simply predict which words usually appear together. If similar combinations featured in their training data, they can confidently state that a date falls on the wrong day of the week.
- What does 'next Tuesday' mean to an AI assistant?
- It depends on how the model is configured, but it will typically interpret it rigidly based on the date it processes the text, without grasping whether the original sender meant the current week or the following one.
- How can I prevent an AI from scheduling meetings at odd hours due to time zones?
- Always check that the original message contains an explicit time reference such as GMT, UTC, or a reference city, and verify the converted time before accepting the invitation in your calendar.