← AI for Real Estate Pakistan
0/15 complete

Module 03 · Property Valuation Research

Building a Comparable Sales Sheet With ChatGPT

25 minfocused lesson7practical steps4grounded questions4source links
Open lesson + course map

On this lesson

Course outline

A comparable sheet organizes properties similar enough to inform a pricing conversation. It does not turn portal advertisements into completed sales, and ChatGPT cannot certify market value. The value of the sheet is the visible inclusion logic, source trail, and arithmetic.

After this lesson, you can build a comparable-listing sheet, normalize units, calculate asking price per area, and explain why each record is included or excluded.

// concept

Define the Subject Before Collecting Comparables

Create a subject card:

// prompt — copy me4 lines
valuation_date | city | exact locality/block | property_type |
source_area | verified_area_basis | beds/baths/floor if relevant |
construction_age/condition | possession/document state |
orientation/access/features | known defects | unknowns

Write inclusion rules before seeing prices. For a flat, they might be: same block or defensible adjacent micro-market, same property type, area within a stated band, similar bed count, comparable floor/lift condition, and current listing observed within a chosen date window.

The misconception is that more rows make an estimate stronger. Ten mixed rows can be worse than four clearly comparable records. Exclusion reasons are evidence, not wasted work.

// concept

Build the Sheet and Preserve Source Values

Use these columns:

// prompt — copy me5 lines
comp_id | source_url | observed_at | listing_status | city | locality |
property_type | source_area | source_unit | conversion_basis |
normalized_sq_ft | asking_price_pkr | asking_price_per_sq_ft |
beds | baths | floor | condition | possession_state | key_differences |
duplicate_group | verification_state | include | exclusion_reason

Formula:

// prompt — copy me1 line
asking_price_per_sq_ft = asking_price_pkr / normalized_sq_ft

Keep the source area and conversion basis beside the normalized value. If the scheme-specific area basis is uncertain, do not normalize it silently. Use Zameen’s current converter as a convenience check, then verify the property’s actual dimensions or scheme convention.

Detect duplicates using contact, photos, exact wording, location, size, and price—but do not publish private contact details in the sheet you share. Keep one representative record and note the duplicate group.

// concept

Let AI Classify, Not Invent Adjustments

AI can compare categorical fields and draft questions:

// prompt — copy me5 lines
Compare each candidate row with SUBJECT and INCLUSION RULES. Return comp_id,
include (yes/no/review), exact matched fields, material differences, missing
facts, and exclusion reason. Do not estimate value, sale price, appreciation,
or adjustment percentages. Treat every portal price as advertised asking price.
Preserve row IDs.

Review the classification manually. Never ask the model to decide “corner adds 10%” or “new construction adds 15%” without a defensible, current local evidence method. Record qualitative differences and obtain a qualified appraisal when the decision requires one.

// concept

Summarize Without False Precision

For included advertised comparables, calculate count, minimum, maximum, median asking price, and median asking price per normalized area. Median is the middle value after sorting and is less distorted by one extreme listing than the average. It is still only a statistic about your selected advertised sample.

Always display:

  • observation window and source;
  • exact inclusion/exclusion rules;
  • count after duplicate removal;
  • missing facts;
  • subject differences;
  • no completed-sale evidence unless a legitimate source actually provides it;
  • no legal/title conclusion.

// worked_example

Worked Example

IDArea sq ftAsking PKRPKR/sq ftDecision
C11,18023,600,00020,000include
C21,22025,010,00020,500include
C31,20027,600,00023,000review: renovated
C41,75030,625,00017,500exclude: materially larger/3-bed
C51,20024,000,00020,000exclude: duplicate of C1

Arithmetic check: 23,600,000 / 1,180 = 20,000; 25,010,000 / 1,220 = 20,500; 27,600,000 / 1,200 = 23,000.

Using C1 and C2 only, the midpoint/median for two values is (20,000 + 20,500) / 2 = 20,250 PKR/sq ft. The report does not multiply that by subject area and call the result “market value.” It states that two non-duplicate asking listings matched the strict rules, which is too thin for a confident valuation. C3 is shown separately because renovation evidence needs review.

// failure_cases

Failure Cases to Diagnose

7 cases to diagnose

  • Completed-sale language is used for portal ads

    relabel all values as advertised asking prices.

  • A duplicate survives because the URL differs

    compare photos, wording, property facts, and restricted contact data.

  • Area conversion has no basis

    preserve source units and mark normalization unresolved.

  • Mean is reported without distribution

    show count, range, median, and every included row.

  • AI creates adjustment percentages

    remove them and retain qualitative differences.

  • Localities or property types are mixed

    enforce the prewritten inclusion rules.

  • A thin sample becomes a precise valuation

    state insufficiency and seek stronger evidence or a qualified professional.

// pakistan_angle

Pakistan Angle

Pakistani portal prices may be written as lakh/crore, while calculations need integer PKR. Convert carefully: one crore is 10,000,000 PKR and one lakh is 100,000 PKR. Check every conversion twice and keep the original string. Also preserve square-yard, square-foot, marla, or kanal source units; local conventions can make a silent marla conversion materially wrong.

A comparable sheet does not verify ownership, society approval, dues, transfer status, possession, or tax. Those questions belong to the relevant authority, society, land-record route, and qualified legal/valuation professionals. Redact seller phone numbers, CNICs, and ownership files before using AI or sharing the sheet.

// hands_on

Hands-On Exercise

7 steps

  1. Create one detailed subject card and written inclusion rules.

  2. Collect eight labelled sample advertised listings with URLs and dates.

  3. Normalize only values with a documented basis.

  4. Detect duplicates and record exclusion reasons.

  5. Run the classification prompt and review every decision.

  6. Calculate price per area by hand for two rows and compare with spreadsheet formulas.

  7. Write a one-page summary with sample limits and unresolved checks.

// completion_rubric

Completion Rubric

6 checks — tick as you verify

0/6

// sources

Sources

4 official sources — check every claim yourself

// check_yourself

Check yourself

4 questions · answers and options are taken word-for-word from this course

0/4

  1. 1 / 4 · diagnose

    Your work shows this failure mode: “A duplicate survives because the URL differs.” What does the lesson tell you to do about it?