REALTOR AI PLAYBOOK · 27
Better Marketing. Stronger Conversations. Practical AI.
Your newest post received fewer likes. That is an observation, not a complete diagnosis. Before asking ChatGPT why your Instagram “isn't working,” identify what the post was supposed to help someone do and which evidence could answer that question.
Make a three-column note now: what happened, what it might mean, what to check next. Put only measured facts in the first column. Keep explanations in the second. Make the third a small, reversible decision.
This chapter gives you a clearly synthetic dataset, checked calculations and a reusable analysis table. None of the numbers are Magic Lens results, client results or a live Instagram export. The exercise teaches a reasoning process without pretending that four invented posts reveal an algorithm.
Choose the question before choosing the metric
A seller-preparation carousel and a property tour may serve different purposes. The carousel might help people retain a checklist. The tour might help someone understand the layout and decide whether to ask a property question. Ranking both by one number can hide that distinction.
Write one primary objective for each content item. Examples include explaining a preparation task, answering a common question or inviting a specific inquiry. Then identify a relevant observable signal. A save can support a question about retained content; it cannot establish that the person hired you.
Separate platform activity from business outcomes. A conversation recorded in your own approved process is different from a content interaction. Do not label every message a qualified lead, and do not count a sale as attributable to a post merely because the buyer follows your account.
Keep the comparison reasonably consistent
Compare posts with a similar objective, format and observation window. Record whether distribution was organic or paid and whether the audience or offer changed. These notes do not remove every difference, but they make the limitations visible.
Preserve the exact metric names, definitions and dates supplied by your current account or export. Labels can change. Do not ask ChatGPT to silently treat views, accounts, interactions and inquiries as interchangeable quantities. If you cannot establish a definition, mark the field as unresolved.
For a beginner, two comparable posts are enough to practice careful description. They are not enough to establish a reliable universal rule. More rows do not automatically fix inconsistent definitions or missing context.
A synthetic dataset you can calculate by hand
The following four invented educational carousels share the same fictional objective: help sellers retain a preparation checklist. Each is observed for its first seven days, with no paid distribution in the scenario. “Accounts” is the exercise's defined exposure denominator; these are not screenshots or claims about the current Instagram interface.
Fictional inputs
A: 800 accounts, 16 saves.
B: 1,200 accounts, 18 saves.
C: 600 accounts, 18 saves.
D: 1,000 accounts, 20 saves.
For this exercise only, calculate saves divided by accounts, multiplied by 100. Call the result saves per 100 accounts. It is a descriptive ratio, not a conversion rate or a platform benchmark.
A produces 16 ÷ 800 × 100 = 2.0. B produces 18 ÷ 1,200 × 100 = 1.5. C produces 18 ÷ 600 × 100 = 3.0. D produces 20 ÷ 1,000 × 100 = 2.0.
D has the largest raw save count. C has the largest ratio. Those statements answer different questions. Neither establishes that C or D will generate more seller appointments, and neither proves that a headline, design choice or posting time caused the difference.
Do not let the average hide its denominator
The totals are 72 saves and 3,600 post-level account counts. Dividing 72 by 3,600 gives a pooled descriptive ratio of 2.0 saves per 100. Adding the four ratios and dividing by four gives 2.125. The numbers differ because the second calculation gives every post equal weight regardless of its denominator.
Neither calculation should be relabeled as the percentage of unique people who saved something across the campaign. The same person could appear in more than one post's account count. Summing those counts does not deduplicate people.
If a denominator is zero, the ratio is undefined; do not replace it with zero percent. If a value is missing, label it missing rather than assuming no activity. Keep the raw counts beside any calculated ratio so the reader can inspect the arithmetic.
Separate observation, interpretation and decision
Analysis table — synthetic exercise, not Magic Lens results
Observation
C has 18 saves from 600 accounts: 3.0 per 100. B has the same 18 saves from 1,200 accounts: 1.5 per 100.
Possible interpretation
C may have been more useful to its audience, or distribution and audience differences may explain the ratio. The data does not identify the cause.
Next decision
Review the content and test one specific opening in another comparable carousel. Keep the objective and seven-day window similar; record the change and limits before reviewing the result.
That decision is stronger than “post more like C” because it names the hypothesis and the next observation. It also avoids turning an interesting ratio into a confident claim about all future content.
Ask ChatGPT for analysis, not a verdict
Analyze this clearly synthetic Instagram content exercise. Do not describe these numbers as real account or Magic Lens results. Objective: help sellers retain a preparation checklist. Format: educational carousel. Window: first seven days per post. Distribution: organic only in this fictional scenario. Defined exercise inputs: A: accounts 800, saves 16 B: accounts 1200, saves 18 C: accounts 600, saves 18 D: accounts 1000, saves 20 Calculate saves / accounts * 100 for each row. Show the arithmetic. Keep raw counts beside the ratios. Calculate the pooled ratio and the unweighted mean separately. Explain why they differ. Do not treat summed account counts as unique campaign people. Return a table with Observation, Possible interpretation, and Next decision. Label explanations as hypotheses. Identify missing context and propose one small comparison to run next. Do not infer revenue, appointments, causality, algorithm preferences or statistical significance from this exercise.
Check the arithmetic independently before using the answer. A well-formatted table can still contain a wrong denominator, a fabricated input or an unjustified conclusion. The expected calculations above make this exercise auditable without trusting the assistant.
A filled decision record
Proposed next comparison—not a tested result
Question: does a more specific opening help the next seller-preparation carousel communicate its purpose?
Change: replace a general opening with a clearly named preparation problem. Keep the content objective, format and seven-day review window similar.
Record: the exact opening, distribution notes, raw counts and the same ratio. Review the quality of any relevant questions separately.
Limit: this is an observational comparison, not a randomized experiment. Differences in topic, timing and audience may still affect the result.
Do not change the opening, format, offer, distribution and measurement window at once, then credit the opening for the result. If practical circumstances require several changes, document them and make a narrower claim about what you learned.
Two refinements that improve the review
Audit your analysis. List each factual statement, its source row and any calculation used. Flag any sentence that presents a hypothesis as a proven explanation. Do not add missing measurements.
This pass checks whether the answer stays connected to the supplied evidence. It is especially useful when the first response sounds more certain than the input deserves.
Reduce the next-step plan to one content decision I can actually execute. Specify the objective, proposed change, review window, measurements and conditions that would leave the result inconclusive.
The second pass turns an analysis into manageable work. If the result is inconclusive, preserve that conclusion. Do not force every review to produce a success story.
Your reusable review table
CONTENT ID / PURPOSE / FORMAT: SOURCE, EXACT METRIC DEFINITIONS AND DATE WINDOW: RAW COUNTS AND DISTRIBUTION NOTES: CALCULATION AND DENOMINATOR: OBSERVATION: POSSIBLE INTERPRETATION: MISSING CONTEXT: NEXT DECISION, OWNER AND REVIEW DATE:
Use a small sanitized table when working with real account data. Remove usernames, private messages and unnecessary contact information. If you do not use AI, fill the same fields manually and check the calculations with a calculator.
Your next action
Select two posts with the same purpose. Record their definitions and equal observation windows. Write one fact, one hypothesis and one next decision before deciding whether the content “worked.”
Turn the finding into a production question
You now have a fact, a hypothesis and a next decision. If the review points to a visual question—such as a room relationship that is difficult to follow—write down the view you would need to examine it. Do not assume that a lower metric automatically calls for a new shoot.
When new coverage has a clear purpose, Magic Lens services and pricing can help you plan photography or video for a Central Florida property.
