Gains:
- Understanding that developmental screening (such as Denver II, AGTE, GEÇDA) is not for diagnosis, but for risk determination and guidance, and positioning the role of artificial intelligence within this limit
- Use artificial intelligence to produce a clear preliminary summary and report draft for the screening results and be able to make the final evaluation and referral decision as an expert
- Ability to catch traps such as age correction, cultural appropriateness, single session error and cut-off score before interpretation
A family comes to you with concerns about their two-and-a-half-year-old child: "His friends talk, but mine talks less." Behind this sentence sometimes lies a completely normal individual difference, sometimes a developmental delay that needs to be supported, and sometimes a situation that requires further evaluation. Separating these three is one of the most critical jobs of the child development specialist. The first step in making this distinction is developmental screening. In this unit, we will cover what screening is, what it is not, and where AI can be placed in the process — with great care.
First the most critical sentence: Developmental screening is not a diagnosis. Screening is a short and practical screening process that helps quickly identify children at developmental risk, further evaluate them, and direct them to the appropriate specialist/physician. Diagnosis is the job of detailed clinical evaluation, multiple data sources and the relevant physician/expert team. If you do not internalize this limit once, artificial intelligence will repeatedly suggest you sentences that exceed this limit, and you will be able to accept them without realizing it.
Scanning tools and basic concepts
Some of the frequently used developmental screening and evaluation tools in Türkiye are: Denver II (0-6 age developmental screening test; scans personal-social, fine motor, language and gross motor areas), AGTE (Ankara Developmental Screening Inventory; 0-6 years, based on family report), GEÇDA (Gazi Early Childhood Assessment Tool; 0-72 months). The common feature of these tools is that they are validated and standardized. "Standardized" means that the tool has been developed on a specific sample and its norms (expected values according to age) have been determined.
Let's clarify a few terms. A norm is a reference value that indicates when most children of a given age have mastered a skill. The cut-off score is the threshold that draws the line "there is a risk / further evaluation is required". A false negative is when a child who is actually at risk passes the screening “normally” — the most dangerous mistake because support is delayed. A false positive is actually marking an otherwise normal child as “at risk” — causing unnecessary concern for the family but usually corrected by further evaluation. A good scan tries to keep false negatives low.
Caution: No scanning tools are "run" within the AI. Scoring is done according to the tool's own manual and validation protocol. Do not use AI to calculate scores or set cutoff scores; Use it only to translate validated results into understandable language and draft reports.
The right role of artificial intelligence
Artificial intelligence does not provide the diagnosis or score during the screening process. Its safe job is to: after you apply the tool and obtain the correct scores, to translate those results into a simple and cautious preliminary summary that is suitable for the family and the file. In other words, AI translates the raw result into a draft that humans can understand; You set the clinical substance of the interpretation and the guiding decision.
Here is the safe flow step by step:
- Apply and score the tool according to its rules. This is entirely your business; AI does not interfere.
- Prepare the result in an anonymized and structured format. Instead of name, "3 years and 4 months old boy", field findings.
- Have the AI write only abstracts and drafts. With the limit of "make no diagnosis, add cut score comment".
- Check for pitfalls like age correction, cultural appropriateness, single session effect, etc.
- Make the final evaluation and guidance decision as an expert and sign it.
Stage
who does
Role of AI
Applying and scoring the tool
expert
None
Risk determination based on cut-off score
Expert (tool guide)
None
Translating the result into plain language
AI under expert supervision
sketch generator
Draft report/family summary
AI under expert supervision
sketch generator
Referral and diagnostic decision
Specialist + physician/team
None
Pitfalls to avoid before verification
Age correction. In a baby born prematurely, development is evaluated according to corrected age, not calendar age. Corrected age = calendar age − (40 weeks − week of birth). For example, the corrected age of a 6-month-old baby born at 32 weeks is approximately 4 months. If you skip this correction you will see a normal baby "behind". Correction is usually made until the age of 2.
Cultural and linguistic appropriateness. An item of a tool (e.g., skill related to a particular toy or foreign object) may not have the same meaning in every culture. For a child growing up bilingual, the number of words in one language may seem less than it is; total language repertoire should be evaluated.
Single session fallacy. The child may not show his true level when he is hungry, sleepless, sick or in a foreign environment. A single scan photo is not the entire film; Repetition and multiple sources (family notification, teacher observation) are required when necessary.
Cut-off score blindness. Being just below or above the cutoff score does not create a sharp "healthy/ill" distinction. Near-borderline results are handled with clinical judgment.
Tip: When printing a summary of results, ask the AI to clearly mark borderline or ambiguous areas as "[expert clinical judgment required]". This prevents you from blindly trusting the outline.
Four copyable prompts
1) Prompt that converts the scan result into a simple family summary:
Your role: drafting assistant to the child development specialist. DO NOT DIAGNOSE, do not comment on the cut-off score, do not say "there is a delay". Translate the following anonymous screening findings into a SIMPLE and CAREFUL summary for the family. Don't be accusatory; Also mention strengths; Where necessary, use directive but non-committal language such as "further evaluation may be recommended." Findings: [anonymous, fielded]
2) Age correction calculation and warning prompt:
I will evaluate a baby's development. Calculate CORRECTED age with the information below, show the formula step by step, and remind you what the difference between calendar age and corrected age can change in the assessment. Birth week: [X] weeks. Current calendar age: [Y months Z days].
3) Report draft skeleton prompt:
Create an ASSESSMENT REPORT DRAFT skeleton from the anonymous scan findings below. Sections: General information, Applied tool, Field-based findings, Strengths, Areas to pay attention to, Recommendation/Guidance (cautious), Note: Add the sentence "This is a screening summary, not a diagnosis". Write [let the expert fill in] in each field you left blank. Findings: [anonymous]
4) Trap checklist prompt:
Check the screening summary draft below for the following pitfalls and itemize any problems you find: (1) is there any language that suggests a diagnosis? (2) does it convert the cutoff score into a definitive decision? (3) Has a situation requiring age correction been omitted? (4) has cultural/linguistic relevance been overlooked? (5) Does it present the single session as absolute truth?Draft: [text]
three mini cases
Case 1 — Near-borderline result. The language field screening result of a child aged 4 years and 1 month is just below the cut-off score (for example, cut-off score of 12, child is 11). The AI draft states definitively that "there is a delay in language development." The expert corrects this to say "a near-borderline result was observed in the language area; further evaluation and support with family notification are recommended." The difference: a choice of words changes the family's level of anxiety and the accuracy of the prompt.
Case 2 — Age correction. The motor development of a baby born at 28 weeks and whose calendar age is 12 months is evaluated. Corrected age is approximately 9 months. In the first draft, AI uses calendar age and says "3 months behind in gross motor". The expert applies the correction; The baby is normal for his/her corrected age. Unnecessary referral and family panic are prevented.
Case 3 — Bilingual child. The number of Turkish words of a 3-year-old child who speaks two languages at home appears to be low. The AI draft only looks at the Turkish repertoire and writes "marked delay in expressive language." The specialist evaluates the child's total repertoire and communication intention in both languages; The picture is more positive than expected. The cultural-linguistic context changes the outcome.
Weak prompt / Strong prompt
Weak prompt:
According to this screening result, write a report to see if the child has developmental delay.
Problem: Asks AI for diagnostic decision, converts cutoff score to precision, no trap check, identity/context is ambiguous.
Powerful prompt:
Your role is draft assistant. Turn the ANONYMOUS screening findings below into a family-friendly SUMMARY DRAFT. Making a diagnosis; turning the cutoff score into a final decision; mark areas near the border as [expert judgment]; Remind me if there is something that needs to be corrected. Add a note at the end, "This is a screening summary, not a diagnosis." Results: [anonymous, field by field]
Common mistakes
- Turning screening into diagnosis: Use cautious language, “further evaluation may be recommended” rather than “there is a delay.”
- Forgetting age correction: Corrected age is essential for premature babies.
- Absolutizing the cut-off score: Cut-off values require clinical judgment.
- Considering a single session as an absolute fact: The child's behavior that day is not always typical.
- Having AI calculate scores: Scoring is done expertly, with the vehicle's own protocol.
In summary
Developmental screening determines risk, not diagnosis; This border is the beginning of everything. Artificial intelligence does not score or diagnose during the screening process; It simply translates the verified results into a simple and cautious summary/report draft. Catch pitfalls like age correction, cultural-linguistic appropriateness, single-session effect, and cut-off score blindness before commenting. The final evaluation and guidance decision belongs to the expert with his signature.
Application task
Get an (anonymized) scan result in your hand. First verify the scoring with the vehicle's own manual. Then, have a family summary draft produced with the 1st prompt, and a report skeleton with the 3rd prompt. Then have the draft checked for traps with the 4th prompt. Finally, translate the diagnostic or definitive statements you corrected in the draft into cautious language and note at least two corrections.
checklist
- [ ] I administered and scored the tool according to its protocol; I didn't have the AI score it.
- [ ] I retained the screening-diagnostic distinction; I translated the precise statements in the draft into cautious language.
- [ ] If there is prematurity, I calculated the corrected age.
- [ ] I evaluated cultural/linguistic appropriateness and bilingualism.
- [ ] I handled borderline results using clinical judgment.
- [ ] I anonymized the data; I added the note "scan is not diagnosis" to the summary.
- [ ] I made the final referral decision as an expert and signed it.