Fertility Clinic Intake Form: Digital Version, AI Summary

Build the form in a tool that lets you control who sees the answers, collect only what the first consultation needs, record consent for each use of the data, and then send the answers to an AI tool approved for health data with a short summary prompt. A clinician reads the original answers alongside every summary before the appointment.

The second half of that sentence is the one that matters. A fertility clinic holds some of the most sensitive information a person will ever write down: medical history, relationship details, previous losses, sometimes finances and sexual health. An AI summary can shorten a clinician's preparation for each new patient, but only if the form is designed with restraint and the summary is treated as a reading aid. Blank form templates are everywhere; the summary step, and the checks around it, are what they leave out.

Follow me on Instagram@sagnikteaches

Decide what the first consultation actually needs

Before opening any form builder, take the last ten first-consultation notes your clinicians wrote and underline every fact they used. Anything that nobody underlined is a candidate to leave off the form. This is data minimisation: collect what you need for the stated purpose and no more. It cuts risk, shortens the form and improves completion.

Connect on LinkedInSagnik Bhattacharya

If you have patients in the EU, the GDPR lists data concerning health, genetic data and data about a person's sex life among its special categories (Article 9), and a fertility form can touch all three. Most other regimes treat health data with similar care. That means a lawful basis and an extra Article 9 condition for using it, such as explicit consent for a specified purpose or the provision of health care, a clear purpose, and tight access. Your data protection lead or adviser sets those; the form's job is to make them easy to follow.

Subscribe on YouTube@sagnikteaches

A realistic split for a first fertility consultation looks like this:

The last row is a good example of restraint. Some details are best asked face to face, where the clinician can explain why. Putting them in a web form invites people to skip the form or to feel watched by it.

Designing the form so people finish it

Completion drops when the form feels like an interrogation. A few design choices matter more than any tool.

• Tell people why each section exists. One line above each: "We ask about medicines so your doctor can check them against any treatment."
• Make most answers tick boxes or short choices and keep free text for "anything else we should know".
• Allow "prefer to discuss in person" on every sensitive question, so skipping is a legitimate answer rather than an abandoned form.
• Save and return. Nobody fills in a surgical history in one sitting. Pick a builder that saves progress.
• Keep it to about 15 minutes. If testing takes longer, cut questions rather than shrinking the font.
• Offer a phone or paper route for patients who cannot or will not use a form online.

When choosing a form tool, look for these on its plan: access controls by role, encryption, an audit log, an agreement covering health data, data stored where your obligations allow, and an export or deletion route. A consumer form tool on a free plan is rarely suitable. Ask the vendor in writing whether health data is permitted and keep the reply.

Consent wording that does real work

Consent on an intake form should be specific, one purpose per tick, and separate from the terms of treatment. For a form that feeds an AI summary, use three separate boxes rather than one catch-all:

1. I agree that the clinic may store the information I enter on this form
   to prepare for my consultation. (Required to continue.)

2. I agree that a computer-based tool may be used to draft a short summary
   of my answers for my clinician, who will check it against my own answers.
   (Optional. If you do not agree, your clinician will read the form directly.)

3. I agree that the clinic may contact me by text message about
   this appointment. (Optional.)

The optional second box is deliberate. The GDPR's test for freely given consent (Article 7(4)) asks whether a service is made conditional on agreeing to processing it does not need, and the clinic can prepare for a consultation without an AI summary. If declining it stops a patient proceeding, it is not a real choice. The same article lets a patient withdraw consent at any time, and withdrawing must be as easy as giving it, so add a line on how to switch the summary off later. Build the process so that a "no" simply routes that form to a clinician with no AI step. Also tell people what the summary is not: it does not make decisions, and a person reads their original answers.

Wording around the form should say who sees the answers, how long they are kept, and how to ask for access or deletion. Link to the clinic's privacy notice rather than repeating it. For more consent wording that patients actually read, see telling patients you use AI note-taking.

Preparing the answers before AI sees them

The safest summaries start with less data. Before any answers go to an AI tool, strip what the summary does not need.

1. Remove direct identifiers. Replace name, date of birth, address and phone number with a case code such as "Case 0417". Keep a separate lookup that never goes into the tool. A coded record is pseudonymised, not anonymous: the GDPR still treats it as personal data because the lookup can re-identify it, so the approved-tool rule below still applies.
2. Drop free-text fields you do not need summarised, such as funding details.
3. Keep the partner's answers apart unless the clinician needs both in one summary and both consented.
4. Send only through the route you approved, never by pasting into a personal account on an unapproved tool.

Which tool? For a clinic, look at business or enterprise tiers that state they do not train on your content by default, offer a data processing agreement, and let your administrator control retention. Facts differ by vendor and change, so read the vendor's current privacy and security pages yourself and ask for a written confirmation covering health data. The wider question of what happens to pasted data is covered in is it safe to put customer data into ChatGPT, and the checklist in patient data and AI goes item by item.

The summary prompt, with a sample result

A good prompt tells the tool exactly what shape to produce, forbids guessing, and keeps the clinician in charge. This is a template to adapt:

You are helping a fertility clinician prepare for a first consultation.
Below are one patient's intake answers, identified only by a case code.

Write a summary of no more than 180 words with these headings:
1. Reason for visit
2. Relevant history (conditions, medicines, allergies, surgeries)
3. Previous fertility care
4. Lifestyle points the patient mentioned
5. Questions the patient wants answered
6. Gaps: questions left blank or answered "prefer to discuss"

Rules:
- Use only what is written in the answers. Do not infer, diagnose or suggest treatment.
- Quote the patient's words in quotation marks for anything emotional or unclear.
- If an answer is ambiguous, list it under Gaps instead of interpreting it.
- Flag in a line at the top, headed ATTENTION, any mention of self-harm, abuse, severe pain or heavy bleeding.

Answers:
[paste the de-identified answers here]

Here is an illustrative sample of what comes back for a made-up case, with the fixes you would make as the reviewer.

ATTENTION: none flagged.

1. Reason for visit: Trying for 14 months, wants to understand "why it isn't happening".
2. Relevant history: Irregular cycles. Takes a thyroid medicine daily. Allergy: penicillin.
   One past surgery (appendix).
3. Previous fertility care: Blood tests at a GP, results not attached.
4. Lifestyle: Does not smoke. Alcohol "a few drinks a week".
5. Questions: Wants to know whether tests are needed before treatment and the likely costs.
6. Gaps: Partner section not started. Cycle length left blank.

As the reviewing clinician, you would notice that the summary says "a GP" while the patient wrote only "my doctor", a small invention. You would also open the original answers to confirm the thyroid medicine name, because a summary that drops a dose is worse than none. The remedy is to tighten the prompt ("never add a detail that is not written") and keep the reading-alongside rule.

Where the summary breaks, and how it shows

Three failures are common enough to design against.

Softening a hard sentence

A patient writes: "I lost the last pregnancy at 11 weeks and I don't think I can go through that again." A summary might compress this to "history of one previous loss". The fact is kept; the person is gone. The prompt's rule to quote emotional text exists for this reason. In review, check that any quotation-worthy line has survived with its words intact.

Merging two people's answers

If both partners' answers sit in one paste, the tool may attribute a condition to the wrong person. Summarise each person separately with their own case code and combine them in the clinician's own notes.

Treating a gap as a no

"Allergies: (blank)" can easily become "no allergies" in a smooth summary. The Gaps heading in the prompt makes blanks visible. Run a weekly spot check of ten summaries against originals, and write down every discrepancy. Two or more in ten means the prompt or the tool needs changing before it touches more patients.

The workflow end to end, with timings

Here is a working flow for a small clinic seeing 25 new patients a week, with illustrative timings you can adjust.

1. Patient books. The booking system sends a link to the form with a short plain-English note on what it is for. Automatic.
2. Patient completes the form, ticking the three consent boxes. About 15 minutes of their time.
3. Reminder. If incomplete 72 hours before the visit, one gentle text. See patient reminder wording and consent.
4. Admin check. A named team member checks completeness and clears the form for summary. Two minutes.
5. De-identify and summarise. Approved tool, standard prompt. Under a minute for the tool; the paste takes two.
6. Clinician review. Reads the summary, then the originals for anything flagged or unclear. About five minutes against perhaps fifteen reading the raw form.
7. Store. Originals in the record; the summary kept only if the policy says so, labelled as a working note.

On those illustrative numbers, saving perhaps eight minutes a patient across 25 patients is over three hours a week of clinician time. That is the case for the project, and it only holds if the review step stays. If review quietly shrinks to a glance, you have bought speed with risk.

Rewriting the awkward questions

The wording of a question decides whether a person answers honestly. Three before-and-after rewrites show the pattern, and you can apply it to every question on your own form.

First, a question that carries judgement. Before: "Have you had any lifestyle problems that might affect fertility?" After: "Do you smoke, vape or use recreational substances? You can choose to discuss this in person." The second asks for facts, names no problem and gives a way out. Second, a question that is too broad for a form. Before: "Please describe your medical history." After: "Tick any that apply" over a list of eight common conditions, followed by one short box for "anything else". The summary tool handles ticks reliably and handles essays unevenly. Third, a question that assumes a shape of family. Before: "Husband's details." After: "Partner's details, if you have a partner attending". Small wording changes like this also make the form usable for single patients, same-sex couples and people using a donor, and clinics that overlook this lose people at the first page.

Test every rewritten question by reading it aloud as if you were sitting across the desk from the patient. If you would hesitate to say it in person, soften it or move it to the consultation.

Language, access and people who need help

A digital form leaves some patients behind unless you plan for them. Check the form on a phone first, because many patients will fill it in on one while travelling or in a break at work. Buttons should be large enough to tap, text readable without zooming, and the progress bar honest. For patients using screen readers, make sure every field has a proper label rather than text sitting in an image, and that the form can be completed with a keyboard alone.

If your patients speak several languages, decide whether you will offer translated forms or an interpreter at the appointment. Machine-translating a medical history form is risky because a mistranslated question produces a wrong answer that then feeds the summary. If you translate, have a fluent speaker who knows the clinical terms check each version. Where that is not possible, offer a phone call to complete the form instead, and record that the AI step was not used for that patient if the answers were taken in another way.

Some patients will want someone to help them fill the form in. Allow it, note in the audit line that help was given, and make sure the consent box is ticked by the patient themselves.

What the project costs, as an illustrative sum

A small clinic can price this in a few lines. Say, with made-up round figures, that the clinic pays $40 a month for a form builder with access controls, $30 a user a month for two clinician seats on a business AI plan, and spends one afternoon, about four hours of a manager's time at $30 an hour, building and testing the form. Set-up time is $120 once; running costs are $100 a month.

Against that, if each of 25 weekly new patients saves a clinician eight minutes, that is 200 minutes a week, or about 13 hours a month. At a clinician's cost of $60 an hour the saving is around $780 a month, before counting the benefit of better-prepared consultations. Cut the saving in half to allow for review time and incomplete forms and the project still clears its cost several times over. These figures are made up for illustration; use your own rates. The decision that matters is not the arithmetic but whether the data protections are in place before you start.

Training the team to run it

Three roles touch the process and each needs about thirty minutes of instruction. The front-desk team needs to know how to send the form link, what to say when a patient asks "who reads this?", and how to route a patient who declines the AI box. The admin lead needs to run the completeness check, de-identify consistently and keep the audit sheet. The clinicians need to know exactly how the summary is built, what it can get wrong and what they must still read. Give each person a one-page card with those steps, and set a review date after the first month to change the prompt, the form or the process based on what actually went wrong.

A good first-month review asks five questions. How many forms were completed? How many patients declined the AI step? How many discrepancies did spot checks find? How long did clinicians spend on review? What did patients say about the form? Use the answers to decide whether to continue, narrow the process or stop.

Retention, access and what to delete

Decide on three things before launch. How long the form answers stay in the record, who can open them, and what happens to drafts. Summary drafts and tool chat histories should not pile up in an AI account. Set the tool's retention to the shortest your administrator allows, and delete working copies on a schedule written in your records policy. If a patient asks for their data, you must be able to find the form answers, the summary if kept, and any copies; fewer copies make that simpler.

Keep an audit line on each case: who completed the form, when consent was given, whether the AI step ran, which clinician reviewed. A short sheet does the job. It lets you answer a patient or a regulator without guessing. Filled in for one illustrative week, it might read like this:

The second and fourth rows are the ones that prove the system works: a declined box and a phone patient both reached a clinician without anyone being pushed through the AI step. If every row in a month says "Ran", check whether box 2 is really being presented as optional.

Testing before real patients

Create three invented patients, each with a different shape: a straightforward case, a case with a long, emotional free-text answer, and a case with several blanks and a partner section. Run all three through the form, the consent boxes, the de-identify step and the summary prompt. Have two clinicians review the outputs independently and list what they would change. Fix the form and the prompt, then repeat. Do this until both reviewers say they would be content to rely on the summary as a first read.

If you want a general version of this set-up for another type of practice, patient intake automation in a small practice covers the common ground, and the therapist intake form shows how a similar form handles emotional detail. When you are ready to bring in a second use of AI, such as reminders, add it as its own project with its own consent line.

Questions clinic managers ask about AI and intake forms

Can AI summarise a fertility intake form safely?

Only with the right set-up. You need a tool whose terms allow health data, a data processing agreement, no use of your content for training, and a clinician who reads the original answers alongside every summary. Treat the summary as a reading aid, never as a record or a clinical judgement.

Should the summary go into the patient record?

Keep the patient's own answers as the record. If the clinician wants a summary saved, store it as a clearly labelled working note, marked as machine-drafted and clinician-checked, and delete drafts you do not keep. Your records policy decides retention, so agree it before the first form goes out.

Do both partners fill in the same form?

Usually each person completes their own section, because each has separate health history and consent. Send separate links, store the answers separately, and let each person choose what the other can see. A shared form invites people to leave out things they would not want read aloud.

What if a patient writes something worrying in a free-text box?

Set a rule that free-text answers mentioning self-harm, abuse or an urgent symptom are flagged to a named person the same working day, whatever the AI summary says. Do not rely on the summary to catch them; it can miss or soften a sentence.

Further reads

• AI Chatbot for Healthcare: A Small Clinic's Data and Consent Checklist — Data and consent checks for any patient-facing AI.
• AI Scribe Security: Audio, Retention and BAA Vendor Questions — Vendor questions on audio, retention and agreements.
• AI for Mental Health Providers: A Roll-Out Plan for Group Practices — How to roll out AI in stages in a sensitive practice.
• 10 Admin Tasks a Small Clinic Can Hand to AI This Month — Other clinic admin jobs worth handing over.
• Personal Trainer Intake Forms: Build One With AI, Protect Health Data — A lighter intake form with health data limits.
• How a Clinic Can Use AI to Cut Missed Appointments — Reducing no-shows once intake is sorted.
• AI for Fertility Clinics: Admin Jobs Only, and Where It Must Stop — Admin jobs a fertility clinic can give AI, a stop list for anything clinical or emotional, and how to hand a worried patient to a person.
• AI Tools and AI Development: The Complete 2026 Guide — the AI hub, including every tutorial in the AI-for-business series.

Sources: GDPR text, Articles 5, 7 and 9 and Recital 26 (read November 2026); vendor business-plan privacy pages should be checked directly before use.