Yes, AI can draft the paperwork around an emergency lighting test: tidy log rows, a defect list and certificate wording. It can't do the test. A competent person has to run the test, record the real results luminaire by luminaire and sign the certificate, and AI must never supply a duration, a pass or a date it wasn't given.
The awkward part is that readings arrive as scribbles: a time here, "dim" there, a unit that wouldn't light. The risk with AI isn't that it writes badly. It tidies a note into something more certain than the note was. A time of 2 h 10 becomes 2.10, "dim" becomes "pass", and a missed luminaire vanishes without a trace.
Which test you are certifying, and how often it runs
The routine depends on the standard your system is held to, so start there. Contractor guidance on the testing standard published as BS EN 50172 and BS 5266-8 describes two tests that matter for paperwork: a short functional test every month and a test for the full rated duration every year. BSI's own page for that standard lists it as current but under review, and says it covers the required system records and log book as well as servicing and testing. The code of practice alongside it, BS 5266-1, had a clause headed "Certificates and log book" in the contents of its 2011 edition, and BSI now lists a fully revised 2025 edition (published 29 October 2025, replacing the 2016 one). The standards' model layouts sit behind a paywall, so check the current editions if a client or insurer asks for that format.
| Test | How often | What happens | What gets recorded |
|---|---|---|---|
| Functional test | Monthly; larger systems can be done in sections so each luminaire is tested within the month | The supply is switched off briefly; every luminaire and sign is checked as present, clean and working; recharging is confirmed once the supply is back | Date, who tested, result per luminaire or circuit, defects, action |
| Duration test | Annually | The system runs on batteries for the full rated duration (guidance gives 3 hours as an example); the lights must still be lit at the end | Start and end times, result per luminaire, defects, remedial work and retest |
Premises that follow NFPA 101 instead are held to section 7.9, which sets a functional test of at least 30 seconds about every 30 days and, for battery-powered systems, a test of at least one and a half hours every year. Written records are kept for the authority having jurisdiction. Self-testing equipment can take over the monthly test, provided someone still looks at each unit monthly, but the annual test still applies; computer-based self-testing systems have to run it themselves and keep a history of the results. Which edition of the code applies depends on what the authority having jurisdiction has adopted, so check that before you copy any interval.
Treat every figure in this section as a reason to read your own standard, its current edition and any insurer or authority requirement. Standards under review change, and the blank log templates found online disagree with each other on intervals, which is a good reason not to build a certificate from one.

The Complete Excel Course with AI Integration
From formulas and pivot tables to Power Query, DAX, VBA and AI with Copilot, ChatGPT and Claude — hands-on from the first lesson.
By Sagnik Bhattacharya and Virginia Thorn, Primetime Emmy Award winner
Buy nowA test day: 25 luminaires, three hours, one Saturday
This test day is illustrative: a two-storey office with 25 emergency luminaires and exit signs on its drawing, all self-contained and rated for three hours. The asset list the tester carries has 24 of them, which matters later. On a Saturday, with the building empty, the competent person checks every charge indicator, isolates the emergency lighting supply at 08:00, walks the building at one hour, at two hours and at the end of the three, restores the supply at 11:05 and confirms the indicators come back on. Service contractors time this test for a low-risk moment, phase it in buildings that are always occupied, and allow the batteries time to recharge afterwards.
A unit whose charge indicator is off before the test is unlikely to last it, so that gets noted before the clock starts. By lunchtime the tester has a clipboard sheet and these dictated notes.
sat 8.00 mains off to EM circuits, charge LEDs checked first. 13 (side exit
sign) LED off, nothing when mains off. 9.00 walk all lit bar 13. 10.00 walk, 7
(open office g east) dark, didnt catch when it went. 19 faded around 10.10 when
I was upstairs so 2h10. 21 server room still on at 11 but dim, someones call.
11.00 end of 3 hrs, all other lit. mains back 11.05, indicators back on by 11.20
all bar 13. Tested by [name].
Running the day from one sheet
The log is only as good as the sheet in the tester's hand. One sheet and one routine keep the readings in a form the assistant can tidy without guessing.
- A week before, update the luminaire list from the current drawing, print one sheet per walker with IDs in route order, and tell the person responsible when the building will be on battery.
- At the start, check every charge indicator and note any that are off. Take the start time from one clock that everyone uses.
- Isolate the supply to the emergency luminaires, or use the system's test facility, and confirm each unit has switched to battery.
- At each walk, tick every luminaire as lit or dark. Write a time next to anything that changes while you are looking at it, because memory is a poor source for a log.
- On the last walk, note anything that is lit but visibly reduced, and write who will decide whether that is acceptable.
- Restore the supply, record the time, and check that the charge indicators return. List any that don't.
- Before leaving, photograph the sheets or type them up the same day with the tester's name and the date, and write the next monthly test date on the sheet.
From scribbled readings to a log the AI can't embellish
The prompt does most of the protecting. It fixes the columns, bans decimal hours, gives the assistant a phrase to use when the notes are silent, and reserves anything subjective for the competent person.
You are tidying emergency lighting test notes into a log. Use ONLY the notes and
the luminaire list below. Columns: ID, location, type, lit at 1 h, lit at 2 h,
lit at 3 h, result, note. Rules: write times as hours and minutes, never as
decimals. If something is not in the notes, write "not recorded". Mark Pass only
where the notes say the unit was lit at the end of the rated duration (3 hours).
If the notes say dim, faded, flickering or unclear, write "competent person to
decide". Do not give causes. Do not add luminaires that are not on the list.
Luminaire list: [paste IDs, locations and types]
Notes: [paste]
Even with those rules, the first pass can come back with rows like these (illustrative), and every one of them is a slip.
| ID | Location | 1 h | 2 h | 3 h | Result | Note as drafted |
|---|---|---|---|---|---|---|
| EL-07 | Open office G east | lit | dark | not recorded | Fail | Failed at 2 h |
| EL-13 | Exit sign, side door | dark | dark | dark | Fail | Battery fault |
| EL-19 | Open office F east | lit | lit | dark | Fail | Faded at 2.10 h |
| EL-21 | Server room | lit | lit | lit | Pass | Dim at end |
The fixes are small but they change the record. EL-07 was lit at the one-hour walk and dark at the two-hour walk, so it failed somewhere between the two and the note should say that, not "at 2 h". EL-13's "battery fault" is a diagnosis nobody made; the notes say the indicator was off and the unit never lit, which could equally be a missing permanent supply. EL-19's "2.10 h" is 2 h 10 min turned into a decimal, and 2.10 hours would be 2 h 6 min. EL-21 is marked Pass although the notes leave the call to someone, so it goes back to "competent person to decide" until the person who saw it decides.
The corrected log, its totals and a check against the drawing
After the tester has corrected the four rows and written their own decision against EL-21, the log reads as follows. NM is a non-maintained luminaire, M is maintained and S is a sign.
ID Location Type 1 h 2 h 3 h Result Note
EL-01 Reception entrance NM lit lit lit Pass
EL-02 Reception desk M lit lit lit Pass
EL-03 Ground corridor 1 NM lit lit lit Pass
EL-04 Ground corridor 2 NM lit lit lit Pass
EL-05 Open office G west NM lit lit lit Pass
EL-06 Open office G centre NM lit lit lit Pass
EL-07 Open office G east NM lit dark n/r Fail Lit at 1 h walk, dark at 2 h walk, time not caught
EL-08 Kitchen NM lit lit lit Pass
EL-09 Meeting room 1 NM lit lit lit Pass
EL-10 Meeting room 2 NM lit lit lit Pass
EL-11 Stair A ground NM lit lit lit Pass
EL-12 Exit sign, rear door S lit lit lit Pass
EL-13 Exit sign, side door S dark dark dark Fail Never lit; charge LED off before test
EL-14 Accessible WC NM lit lit lit Pass
EL-15 Stair A landing NM lit lit lit Pass
EL-16 First corridor 1 NM lit lit lit Pass
EL-17 First corridor 2 NM lit lit lit Pass
EL-18 Open office F west NM lit lit lit Pass
EL-19 Open office F east NM lit lit dark Fail Faded about 2 h 10 min
EL-20 Meeting room 3 NM lit lit lit Pass
EL-21 Server room NM lit lit lit Pass* *Dim at 3 h; competent person's decision recorded
EL-22 Exit sign, stair F S lit lit lit Pass
EL-23 First floor WC NM lit lit lit Pass
EL-24 Plant room access NM lit lit lit Pass
Count it yourself rather than trusting a total from any tool: 24 rows, 21 passes and 3 fails. Now give the assistant the drawing's schedule and ask it to compare the two lists without inferring anything.
Compare the drawing schedule with the test log. Do not infer or assume. List
(1) IDs on the drawing but missing from the log, (2) IDs in the log but not on
the drawing, (3) duplicate IDs, (4) rows where the result contradicts the 1 h,
2 h and 3 h columns, and (5) whether passes plus fails equals the number of rows.
Drawing schedule: [paste the 25 IDs and locations]
Log: [paste the log]
1. On the drawing, not in the log: EL-25, first floor store.
2. In the log, not on the drawing: none.
3. Duplicates: none.
4. Contradictions: none. EL-07 fails with dark at the 2 h walk, EL-13 is dark at every walk, EL-19 is dark at 3 h.
5. Totals: 24 rows = 21 passes + 3 fails. The drawing lists 25, so one luminaire has no result.
EL-25 was fitted in a first-floor store after the tester's asset list was last updated. An assistant working from the log alone could never have known it existed. The competent person now decides how to deal with it: test it separately, or add it to a repeat visit. The certificate has to say that 24 of 25 luminaires were tested and name EL-25 as outstanding, and the asset list needs updating so the next test day starts from the right 25.
A defect list the client can act on
Three failures need a list the client can act on and an engineer can price. The assistant can write the layout and the plain-English sentence. The cause is a question for whoever goes back with a ladder, so tell it to label causes as hypotheses.
| Ref | Luminaire | Finding | Likely cause (to confirm) | Action |
|---|---|---|---|---|
| D1 | EL-07, open office G east | Lit at the 1 h walk, dark at the 2 h walk | Battery nearing end of life | Replace battery, then repeat the duration test on this unit |
| D2 | EL-13, exit sign, side door | Charge indicator off before the test; never lit | No permanent supply to the charge circuit, or a failed unit | Check the supply, replace the unit if needed, retest |
| D3 | EL-19, open office F east | Lit for about 2 h 10 min | Battery capacity low | Replace battery, then repeat the duration test on this unit |
The "to confirm" heading is deliberate. A cause stated as fact on a page that will be signed invites a repeat of the EL-13 slip. Each action also ends in a retest, because a repaired luminaire hasn't passed until it has been shown to run for its rated duration.
The certificate: AI drafts the page, the competent person declares
Layouts vary with the standard, the client and the insurer, so this table lists the content rather than a form. Drop each item into whichever layout you are required to use. Every example entry is illustrative.
| Field | Example entry | Where it comes from |
|---|---|---|
| Premises and responsible person | [premises name and address]; [responsible person] | Job record |
| System | 25 self-contained luminaires and exit signs on the drawing, rated duration 3 hours | Drawing schedule, checked on site |
| Test carried out | Annual full-rated-duration test on [date]; supply isolated 08:00, restored 11:05 | Log |
| Extent tested | 24 of 25 luminaires; EL-25 (first floor store) not yet tested | Drawing cross-check |
| Result | 21 lit at the end of 3 hours (EL-21 with reduced output, accepted by the competent person); 3 did not (EL-07, EL-13, EL-19) | Log totals, counted by a person |
| Defects and action | D1 to D3 as listed; retest on [date] | Defect list |
| Conclusion | Written by the competent person | Human only |
| Next tests | Monthly functional tests; next annual duration test due [date] | Competent person |
| Declaration | Name, competence statement, signature, date | Human only |
Two rows are never drafted. Whether a system is satisfactory with three open defects and one untested unit is a professional judgement, and different layouts use different words for it. The declaration works the same way: a signature says a named person ran or supervised the tests and stands behind the record, and an assistant can do neither.
The monthly entry that takes two minutes
The same discipline works for the short monthly test, where the temptation is to write less. Suppose the tester dictates this after a walk-round:
monthly done tues, everything fine except the kitchen one flickers
Run through the same rules, the dictation comes back as a log entry (illustrative): test type, monthly functional; result, all luminaires and signs operated except one; observation, kitchen luminaire flickering, ID not recorded; action, none recorded. The assistant rightly wrote "ID not recorded" and didn't guess which kitchen unit it was, and it also didn't turn "everything fine" into "24 of 24", because no count was given. Both gaps go back to the tester before the entry is filed, since a flicker with no ID can't be fixed.
The quarterly read that catches what nobody chased
Take the manager of a childcare nursery who has kept the monthly log all year. Pasting it into a chat each quarter and asking three questions does more than a reminder email.
Here is the monthly emergency lighting log for [premises] for this year. List
(1) months with no entry, (2) entries that mention a defect but record no action,
(3) any luminaire ID that appears in two or more entries. Do not comment on
compliance.
1. No entry: May and June.
2. Defect without action: March, "hall light slow to come on".
3. Repeated ID: EL-04 appears in February, March and April.
The third finding is the useful one. A luminaire that keeps turning up is probably a failing unit rather than three separate glitches, so it goes to the competent person rather than into the next monthly note. The first finding is a phone call to the manager. The second is a call to whoever was told about the hall light.
Four slips when you reuse last year's paperwork
- Carried-forward results. Paste last year's log and ask for an update, and rows this year's notes don't mention can inherit last year's passes. Always draft from this year's notes into an empty table.
- The template's duration. A blank certificate says three hours and your system is rated for one, or the reverse. Rated duration comes from the design documents or the luminaire labels, so state it in the prompt yourself.
- An ageing asset list. The EL-25 problem. Draft from the current drawing, not the list from three years ago.
- A recycled declaration. Last year's wording carries last year's date and competence statement. Re-enter both when you sign.
Routines for a nursery, a care house and a warehouse
A childcare nursery that closes at weekends
The building is empty on Saturdays, so the annual duration test has a natural home, and the monthly walk can happen on the same weekday before opening. The manager is the person responsible and keeps the log in the office folder. Here the assistant earns its keep on chasing, through the quarterly read above, and it helps if room leaders note any unit that looks wrong between tests in the same log, with the ID from its label, so the next read picks it up. The competent person still runs the annual test and signs. A reasonable trigger is two empty months in a row, which means a phone call rather than a polite reminder.
A care agency's residential house
People sleep here, so the whole building shouldn't sit on battery at once. Phase the annual test by zone, as contractors do for permanently occupied buildings. Ask the assistant to group the luminaire list by floor into zones of a sensible size and add a Zone column to the log, then have the competent person approve the plan and the timing with the duty manager. Each zone gets its own start and end times and its own recharge check before the next zone begins. Because the test is spread over several days, the certificate needs the date of each zone's test rather than a single date, which is easy to lose in a draft.
A craft brewery supplier's warehouse
High-bay units, exit signs mounted well above head height and racking that gets rearranged: the test itself is simple but the access isn't. An assistant can order the walk route by aisle and add an "access equipment" column from the asset list, writing [CHECK] where mounting heights aren't recorded. Whether a rearranged rack now blocks a luminaire's light is a different matter. That is a question about whether the installed lighting still does its job, and it belongs to the competent person, who may want light readings. A warehouse log should therefore have an observations column that the assistant never fills in.
What stays with the competent person
- Running the test and looking at each unit with their own eyes, at each walk.
- Judging anything that isn't a plain fact: dim, flickering, partly obstructed, discoloured.
- Deciding whether coverage and light levels still suit the building, which no log can answer.
- Deciding what the conclusion says when defects are open or a unit is untested.
- Telling the person responsible for the premises what must happen next, and by when.
- Signing, dating and stating their competence.
The test that settles every doubtful row is this one: a line in the log that didn't come from someone looking at that luminaire isn't a record. Used that way, an assistant saves the typing, the tallying and the chasing, and the evidence stays human. If you also run electrical certificates, the electrical installation certificate tutorial draws the same line for a different form, and fire door inspection reports do the same for another competent-person check.
Once the paperwork is right, the next problem is remembering it. Annual and monthly dates drift, and a compliance calendar with reminders sent to the right person is cheaper than a missed test. If you service several sites, AI for inspection reports and reminders covers the asset list and due-date side, and inspection logs for facilities managers covers the same job from the building owner's chair.
Emergency lighting paperwork, answered
Can AI mark a luminaire pass or fail from my notes?
Only when the note states the fact, such as 'lit at the end of three hours'. Words like dim, faded or flickering need a person's judgement, and so does any decision about whether light output is adequate. Tell the assistant to write 'competent person to decide' for anything that isn't a plain fact, and read those rows yourself.
Does a self-testing system still need a duration test?
Self-testing equipment automates the short monthly test, although NFPA 101 still expects a visual check of each unit every month. It still calls for the annual test of at least one and a half hours on battery-powered systems, and computer-based self-testing systems must run that test automatically and keep a history. For other standards, check the standard and the manufacturer's instructions rather than assuming the automatic test replaces the annual one.
How long should the batteries be left to recharge after the annual test?
Follow the manufacturer's stated recharge time and your standard. Service contractors time the annual test for a quiet period so the batteries can recharge, and phase it in buildings that are always occupied. Before leaving, check that the charge indicators have come back on, and record any that haven't.
What should the log hold for each test?
At minimum the date, the type of test, who carried it out, the start and end times for a duration test, the result for each luminaire or circuit, the defects found and the action taken. BSI describes the testing standard as covering the system records and log book, so check its detail against your own template.
Is it safe to paste a client's building details into an AI chat?
Use placeholders such as [premises] and [responsible person] while drafting, and add the real names afterwards. Consumer ChatGPT and Claude plans have a switch for model training in privacy settings, and business plans don't train on content by default. Luminaire IDs and locations alone reveal little, but the premises name and address belong on the final page only.
Further reads
- Can AI Design a Fire Alarm System? What an Installer Must Still Do — Where a competent installer's duties begin when AI helps with fire systems.
- EICR Software and AI: What Can Be Drafted, What You Decide — The same drafting-versus-sign-off line for electrical condition reports.
- Scaffolding Inspection Records: A Digital Form and AI Summaries — A digital inspection form built on the same principle.
- Can AI Write RAMS for Electrical Work? Who Must Sign It Off — Who checks and signs AI-drafted risk assessments.
- Gas Engineer Software: What AI Can Draft and What You Must Sign — What AI can draft and what must be signed for gas records.
- EPC Assessor Software: What to Check First, and Where AI Helps — Check scheme approval, approved-list status and update speed before you pay, then use AI for client emails and wording, never for lodgement.
- PAT Testing Software: Where AI Helps With Reports and Reminders — Test the reminder and report claims in PAT software, then let AI draft customer summaries and chasers while the tester keeps every result and decision.
- AI Tools and AI Development: The Complete 2026 Guide — the AI hub, including every tutorial in the AI-for-business series.
Sources: BSI Knowledge page for BS EN 50172:2004 / BS 5266-8:2004 (checked 1 November 2026); BSI Knowledge pages for BS 5266-1 (the withdrawn 2011 and 2016 editions and the current 2025 edition); testing guidance from two emergency lighting service contractors; a regulator's guidance note quoting NFPA 101 section 7.9.3 (checked 1 November 2026).