Preventive Maintenance Schedule Template: AI Builds It From Manuals

Make a list of every machine, upload each manual, and ask the AI to extract every service task with its interval, the page it came from and the manual's own wording. Then check each line against the original page, convert hour-based intervals to calendar dates using real run hours, and group the tasks into a daily, weekly, monthly and yearly calendar.

No sensors or maintenance software are needed for this. A spreadsheet, the manuals you already have and a day or two of checking will produce a schedule better than the one most small workshops run from memory. The AI saves you hours of reading, but the page-number column is what makes the result trustworthy, because it lets you verify every interval in seconds.

Follow me on Instagram@sagnikteaches

Start with an asset register, because the schedule hangs on it

An asset register is simply a list of every machine and piece of equipment worth maintaining, with enough detail to find its manual and to tell two similar machines apart. Without one, the AI produces a schedule for "the compressor" and nobody knows which of the two compressors that means. Keep it as a spreadsheet with these columns:

Connect on LinkedInSagnik Bhattacharya

• Asset ID. A short code you stick on the machine, such as CNC-01 or COMP-02.
• Description, make and model. Copy from the nameplate, not from memory.
• Serial number and year. Manuals often cover several variants, and the serial number tells you which one you own.
• Location. Bay or room, so the person doing the task can find it.
• Manual file name. The PDF in your shared folder.
• Criticality. A 1 to 3 rating: 1 if its failure stops production today, 2 if it slows you, 3 if you can work round it.
• Typical running hours a week. This turns hour-based intervals into dates.
• Isolation point. Where the power, air or fuel is isolated before work starts, per the manual.

Walk the premises with a phone and take a photo of every nameplate. Ask the AI to turn the photos into register rows if you like, but read the serial numbers back yourself; a misread digit sends you to the wrong manual.

Subscribe on YouTube@sagnikteaches

Illustration: a furniture workshop's first register

A twelve-person furniture maker lists 14 assets: a CNC router, a panel saw, an edge bander, a wide-belt sander, two compressors, dust extraction, a forklift, a vacuum press and others. Seven have PDFs, four have paper manuals, two have none, and one is a second-hand saw whose model the owner has to look up. The workshop scans the paper manuals with a phone app that makes searchable PDFs, finds one missing manual on the manufacturer's site, and flags the other as "no manual: ask supplier". That inventory step takes one afternoon and already shows that criticality 1 is the CNC router and the extraction system.

Extract intervals with a prompt that keeps the page number

Upload one manual at a time. Mixing manuals in one request tempts the model to blend machines. Check the file limits of the tool you use: as of November 2026 Claude reads PDFs up to 1,000 pages but analyses images and diagrams only in PDFs of 100 pages or fewer (longer ones are read as text only), which matters for manuals full of exploded views, and ChatGPT accepts documents up to 512 MB, capped at 2 million tokens a file. For a very long manual, upload just the maintenance and lubrication chapters. A Claude or ChatGPT project that keeps your instructions saves pasting the prompt each time. If the diagrams matter, attach the manual to the chat itself rather than saving it as a ChatGPT project file: OpenAI's help page says embedded images in project files may not be included automatically.

Extract every maintenance, inspection, lubrication, cleaning, calibration and
replacement task from the attached manual for [asset ID, make, model].
Output a table with these columns:
Task | Component | Interval as written | Interval unit (hours / days / weeks / months / cycles / "as required") |
Condition (e.g. "severe duty", "first 50 hours") | Page number | Exact quote (max 25 words) |
Who may do it (operator / trained technician / manufacturer) | Safety warning on that page (Y/N, with the page).

Rules: use only the manual. Do not infer or average intervals. If two intervals appear
for the same task, list both with page numbers. If the manual says "as required" or gives
no number, write NO INTERVAL STATED. If a table is unreadable, write UNREADABLE and the page.
After the table list any task that appears in a troubleshooting section but not in the schedule.

Illustrative output for a rotary-screw air compressor, trimmed to three rows:

The third row is a good sign: the AI refused to invent a number. That task is condition-based, so it goes into the schedule as an inspection ("read the pressure gauge weekly") and not as a date.

Check every line against the original page

This step is what separates a schedule from a guess. It takes about ten minutes a manual if you work from the page numbers. For each row:

1. Open the PDF at the cited page and find the quoted sentence. If the quote is not there, strike the row. Claude's help pages say it counts pages as your PDF viewer numbers them, not by the page numbers printed in the manual, so if the cited page looks wrong, check the printed number a few pages either side before you strike it.
2. Read the sentence either side. Intervals are often qualified: "under normal conditions", "in dusty environments halve this". Add a condition column entry or a note.
3. Compare the number. Models sometimes convert units, for example reading "every 500 h" as "every 3 weeks". The unit must stay as the manual wrote it.
4. Check the table's header. A table headed "initial service" is not a recurring interval.
5. For costly machines, check against the manufacturer's current manual or support page. Manuals are revised, and the copy in your drawer may be older than the machine's last service bulletin.

Pay particular attention to tasks that carry a safety warning. If the manual says only a trained technician or the manufacturer may do it, the schedule must say so, and the operator's checklist must not include it.

Why the "exact quote" column matters

A model can paraphrase accurately nine times and drift on the tenth. Asking for a short quote makes a drift visible, because the quote either exists on the page or not. It also gives you something to show an auditor or a warranty claim handler: the task, the manual's words and the page. If you ever need to defend a skipped or changed interval, that column does it.

Turn hours and cycles into dates

Most workshops have no hour meter on half their machines and do not want sensors. Use the hours per week figure from the register instead. Quick sum for the compressor above: it runs about 30 hours a week, so a 2,000-hour interval arrives after roughly 67 weeks, but the manual also says 12 months, whichever first. The calendar interval wins at 12 months, so the filter task is annual. A second task, an oil change every 4,000 hours, lands at about 133 weeks, so it is a two-and-a-half-year task, and you would write it as every 30 months until a meter reading says otherwise.

Where a machine has an hour meter, such as a forklift, read it at each weekly round and record the figure. The spreadsheet then adds the interval to the last service reading to show the next due reading. The aim is to avoid doing everything on the calendar when some intervals genuinely depend on use. Cycle-based tasks, for example a press's cycles, follow the same logic with a cycle counter or a log of jobs.

A rule of thumb the AI can apply: when a manual gives two intervals, take the shorter in calendar terms. Then note it in the schedule. If you later extend an interval, write the reason and who approved it.

Group the tasks into a schedule people will follow

A list of 140 tasks with different intervals is not a schedule. Group them by frequency, and by who does them, so a person gets a sheet they can finish in one go. Ask the AI to do the grouping from your verified table.

From the verified table below, build a maintenance schedule:
1. Group tasks into Daily, Weekly, Monthly, Quarterly, Six-monthly, Annual and Multi-year.
   Round an interval DOWN to the nearest group (a 5-week task becomes Monthly).
2. Within each group, split into Operator tasks and Technician tasks.
3. For each task give: asset ID, task, estimated minutes (write [ASK] if unknown),
   parts or consumables, isolation needed (Y/N), source page.
4. List condition-based tasks separately as inspections with the trigger.
5. Show total minutes per group, per person type, per month.
Do not change any interval or wording from the table.

Rounding down is a deliberate choice. A task due every five weeks done monthly is a little early. Done every two months it breaks the manual's rule.

What the finished schedule looks like

For the furniture workshop, a trimmed extract of the monthly group (illustrative):

The totals show something useful: the whole monthly group comes to about 3.5 hours, roughly 90 minutes of operator time and two hours of technician time. That is the number to take to whoever signs off the time. It is also a reality check: if the schedule would take 40 hours a month and you have 10, you now know which tasks to move to a contractor or which low-criticality assets to cut to the manufacturer's minimum.

Making it stick: sheets, records and ownership

A schedule on a screen does nothing. Print one tick sheet per frequency group, or make a tab per group in a shared spreadsheet with a date column, an initials column and a notes column. Name an owner for each frequency: the operator closes the daily sheet, the workshop lead closes the weekly one, the technician the rest. Record completion dates, because the next-due date column is a formula (last done plus interval) and shows overdue tasks in red. A weekly ten-minute review of the red cells beats any dashboard.

The notes column is where the schedule starts to learn. "Belt found cracked" on a monthly check three times in a row is the signal to shorten the replacement interval, with the manufacturer's agreement. The same notes, pasted into the AI once a quarter, can give you a list of repeat findings. For a bigger view of what stops your machines, reading downtime logs to find repeat stops is the natural companion.

Two ways the schedule goes wrong

The blended manual. In this illustration, a garage uploads two versions of a vehicle-lift manual, an older one and a newer one with a revised lubrication interval, into one request. The AI lists both intervals without saying which belongs to which, and the shorter one is quietly dropped when the table is tidied. The fix is one manual per request, with the revision date copied into the register. A car repair garage that works from a lift's certification inspection report as well should also treat the inspector's findings as a separate source, not mix them into the manual's intervals.

The interval that sounded sensible. In a second illustration, a food importer's warehouse asks the AI to "suggest a schedule" for a pallet stacker without uploading the manual. The result looks plausible and quotes a weekly inspection that the real manual puts at daily for the brakes. Only the check against the manual catches it. Never let a model supply an interval from general knowledge. Anything it cannot source goes to a manual, a supplier or an experienced operator.

Machines with no manual, and manuals that disagree

Every workshop has an old machine with no paperwork. Do not let the AI fill the gap from general knowledge. Work down this order instead: first the manufacturer's support page or archive, searched by model and serial number; then the supplier or dealer who sold it; then a technician who has serviced the model; and last, an experienced operator, whose routine you write down and mark "operator practice, not manufacturer interval". The AI helps in the last case by turning a rambling voice note or typed account into a tidy table, which you then label honestly in a source column.

Disagreement is the other awkward case. A component maker's manual, for a motor say, may give a bearing grease interval different from the machine maker's manual that houses it. Take the machine maker's figure if it is the shorter, note the other, and ask the machine maker if the two cannot be reconciled. Never settle it by averaging.

Build a spares and consumables list from the same table

The extraction table already holds your parts demand: every filter, belt, grease cartridge and blade the schedule says you will need. Ask the AI one more thing from the verified table: list each consumable, the part number if the manual gives one, the quantity per task, the tasks that use it and the first date it will be needed. Then add your own columns: stock on hand, supplier, lead time in days and minimum stock. Anything with a lead time longer than the gap between the order date and the due date goes to the top of the list. The furniture maker's list turned up a dust-filter set with a six-week lead time due in three weeks, which is the sort of thing that otherwise appears as a machine standing idle. If you buy parts through a supplier system, automating purchase orders and supplier emails can send the order when stock hits the minimum.

A full-year workload sum for the whole register

Once every asset is in, total the minutes. Illustrative figures for the 14-asset furniture workshop, after checking and grouping:

That is about 262 hours a year, of which the daily and weekly rounds take 196 and the monthly and longer rounds about 66. Roughly an hour a day across a twelve-person shop is far easier to justify than "we should do more maintenance", and it shows that the daily round is where most of the time sits. The sums also expose which tasks are worth a second look: if the daily round takes 35 minutes and ten of them belong to the extraction system, ask whether the manual really demands each check daily or whether a weekly one would meet its wording.

Safety wording that must not be paraphrased

Manuals contain warnings about isolating power, releasing stored air pressure, locking out hydraulics and keeping guards fitted. The schedule is not the place to summarise them. Add one column, "isolation and safety", that holds the manual's page number for the relevant warning and the isolation point from the register, and make the tick sheet say "Isolate per manual page X before starting". Ask the AI to flag any task whose manual page carries a warning triangle or the word "danger" or "warning", and treat each flagged line as one that a trained person does under your site's own isolation procedure. If you do not have such a procedure, that is a gap for a qualified safety adviser, not for a prompt.

A short example for a car repair garage

Illustration: a car repair garage with a two-post lift, a compressor, a tyre changer, a wheel balancer, a brake tester and an oil-drain unit builds its register in an hour. The lift manual gives a daily visual check, a monthly safety-latch and cable check, and an annual inspection by a competent person. The tyre changer manual gives a weekly lubrication and an air-line filter drain. The AI's extraction takes 20 minutes in all, the page-by-page check another 40. The result is a one-page daily sheet for the first technician in, a weekly sheet for the workshop lead, and an annual list that goes into the calendar with the booking of the lift inspection. Nobody has to remember anything, and each line carries the manual's page for the inspector who asks.

Does the schedule work? Three checks after a quarter

1. Completion rate. What percentage of due tasks were done on time? Below 80 percent means the load is too big or the owners unclear.
2. Findings. How many tasks found something before it failed? Tasks that never find anything may be too frequent, though only the manufacturer can say that safely.
3. Unplanned stops. Compare breakdown hours with the same quarter last year, using your own log. If a failure occurred that no task covered, add one, with its source, or note that the manual has none.

If you run farm machinery or tools out of a depot, a service log for farm equipment built from manuals follows the same pattern with seasonal use, and a gym equipment checklist from manuals shows the checklist side for a different sector. Think twice before reaching for sensors and prediction: predictive maintenance for a small workshop explains when it pays, and for most small sites a calendar built from manuals comes first.

Questions about building a maintenance schedule from manuals

What if I have lost the manual for an old machine?

Search the manufacturer's support pages with the model and serial number; many keep PDFs online or will email one. If none exists, list the machine in the register with a note and use a similar model's manual only as a starting point, labelled as such. Ask an experienced operator what they do already and write that down.

Can I upload scanned manuals?

Often yes, but accuracy drops with poor scans, small print and exploded diagrams, and a model may skip tables. Check the extracted interval against the page image every time. If a scanned manual matters to a costly machine, retype the maintenance chapter's table by hand and let the AI work from that.

Does following the manual keep my warranty valid?

It usually helps, because manufacturers often require maintenance to their schedule and some ask for records. The exact warranty terms are in your purchase paperwork, so read those, and keep dated completion records for each task. If a machine is under warranty and the terms matter, ask the supplier in writing.

How often should I review the schedule?

Review it once a quarter at first, then twice a year. Look at tasks that were skipped, tasks that found a fault every time and breakdowns that no task would have prevented. Change intervals only with a reason noted next to the change, and do not stretch a manufacturer's interval without asking them.

Further reads

• OEE Calculator in a Spreadsheet: AI Sorts Downtime From Notes — Measure what downtime is costing before and after the schedule.
• AI for an Equipment Rental Business: Setup, Cost and What to Automate — Maintenance tracking for fleets that go out on hire.
• AI Quality Control for Small Manufacturers: Where to Start — Link maintenance to the defects it prevents.
• AI Supplier Management: Track Prices, Lead Times, and Risk — Track spare-part suppliers and lead times.
• How Much Does AI Visual Inspection Cost a Small Manufacturer? — What camera-based checks cost if you go beyond a calendar.
• AI Maintenance Costs: What You Pay After an Automation Goes Live — Budget for keeping the AI side running too.
• CCTV Maintenance Contract: Remote Health Checks and Renewals With AI — A monthly remote health-check routine, an AI-drafted report and a five-step renewal sequence that lets a CCTV firm renew maintenance contracts on evidence.
• AI in Plant Engineering: Breakdown Logs and Maintenance Plans — Turn a messy breakdown log into ranked repeat faults and a draft preventive maintenance plan, with the safety items left to a named engineer.
• CE Marking Technical File Checklist: Draft the Outline With AI — A checklist for building a CE technical file outline with AI, and the checks a competent person still has to make.
• Calibration Log Template: AI Tracks Due Dates for Gauges and Scales — A spreadsheet calibration log with AI-built due dates, status flags and reminder emails, where certificates stay the evidence.
• Daily Production Report: Summarise Counts, Scrap and Stops With AI — Three messy shift sheets become one page: units against plan, scrap by reason, stops by cause, and the one thing the owner must act on.
• MRO Inventory Management: AI-Set Minimums for Maintenance Spares — One factory's spares sheet, built from work orders and a criticality score, with min and max rules and an obsolete-parts check.
• AI Tools and AI Development: The Complete 2026 Guide — the AI hub, including every tutorial in the AI-for-business series.

Sources: the machine manuals you load are the primary source; check each extracted interval against the manufacturer's current manual or support page. Upload limits from Anthropic's help article on documents you can upload to Claude (1,000-page cap, visual analysis up to 100 pages, PDF viewer page numbering) and OpenAI's help article on uploading files to ChatGPT (512 MB, 2 million tokens per document, images in project files).