Can AI help us summarise medical records on a personal injury file?

Yes, for the part of the job that is reading rather than judging. An AI assistant can work through several hundred pages of GP notes, hospital letters and physiotherapy records, pull out the entries that mention a given complaint, and tell you which page each one came from. What it cannot do is form a clinical view, decide what is medically significant, or know that a set of records is incomplete.

What the task actually involves

When a fee earner summarises medical records, several different jobs are happening at once:

• putting entries in date order, across records that arrive in no order at all
• identifying who wrote what: GP, treating consultant, physiotherapist, occupational health
• finding the first mention of each complaint, and the last
• spotting pre-existing conditions and earlier similar episodes
• noting periods of absence from work, prescriptions, referrals and discharges
• noticing what is absent: a referral with no resulting letter, a gap in the GP record, a hospital episode with no discharge summary

Only the first five are reading tasks. The last is a judgement task, and it depends on knowing what records you asked for and what you received.

Where an AI assistant earns its place

The strongest use is targeted questioning rather than "summarise this". A fee earner who asks when the records first mention lower back pain, or which entries refer to a knee, or what the records say about time off work, gets a short answer with page references and can go straight to the pages. That is faster and safer than reading a two page narrative summary that has quietly merged three entries into one.

Second, coverage. A person skim reading four hundred pages at the end of a long day will sample. A tool that reads page by page does not, provided it tells you which pages it could not read. Poor scans, faxed letters and handwritten notes are common in medical records, and a page that was silently skipped is worse than a page that was flagged.

Third, extraction into a working document. Dates, treatment types and absences can be pulled into a draft chronology or schedule that a fee earner then checks and edits. The tool produces the first version; the fee earner decides what stays.

What it cannot do, and should not be asked to do

Causation, prognosis and the significance of a clinical entry are matters for a medical expert. An AI assistant that offers a view on whether an entry supports or undermines a claim is being asked the wrong question, and a summary written in that register will be misleading when it reaches the expert or the other side.

It also cannot tell you that records are missing unless the gap is visible on the face of the papers. If the second hospital never sent its file, the assistant will summarise what it has and say nothing about what it does not have. Completeness remains the fee earner's responsibility: check what was requested, what arrived, and what has been chased.

Abbreviations and handwriting are a real limit. Clinical shorthand is dense and inconsistent, and a misread abbreviation can change the meaning of an entry. Anything load bearing should be read at the page before it goes into a schedule, a letter of instruction or a pleading.

Checking the summary before it goes anywhere

A short, repeatable check is more use than a vague instruction to review carefully:

1. Paginate the records first, and record which version of the bundle was used. Summaries that reference unpaginated files cannot be verified later.
2. Verify the anchors: the earliest entry, the latest entry, the date of the index accident and the date of any surgery.
3. Check every date and figure that will appear in a schedule, a letter of instruction or a witness statement against the page it came from.
4. Read the source pages for any entry that cuts against the client's account. These are the ones that will be tested.
5. Note on the file who checked the summary and when.

If the tool says the records do not address something, treat that as a prompt to check whether the records were ever obtained, not as a finding.

Health records and confidentiality

Medical records are special category data under UK GDPR, and they usually concern someone who is not your client as well: treating clinicians are named throughout, and family history entries mention third parties. Before any records go into a tool, be satisfied about where the data is held and processed, whether it is kept separate from other firms' material, whether it is used to train anything, and who inside your own firm can see the matter. The Information Commissioner's Office expects organisations to be able to answer those questions about health data without hesitation, and a client is entitled to ask them too.

It is also worth deciding, as a firm, whether record summarisation is something you disclose in your engagement terms. Many firms cover it with a general statement about using technology under supervision; the point is to have decided rather than to improvise when asked.

Where Alesis fits

Alesis answers questions about a matter from the matter's own papers and names the page each answer came from; if the papers do not say, it says so. Documents are read page by page, so citations point at pages, and any page it could not read is flagged rather than skipped. It prepares drafts for a qualified person to review and sign off, and the firm's information is held in the UK, kept apart from every other firm, and never used to train anything for anyone else. It assists qualified professionals and does not replace them, and it does not provide legal advice.