Your Email Says Optional. Your Text Says Required.

Your Email Says Optional. Your Text Says Required.

Anna with Oppy
November 13, 2026

An email says a service photo is optional. The text says, “Send a photo so we can proceed.” The voice reply politely asks the customer to upload one before scheduling.

Same business. Three channels. Two invented requirements.

A channel can sound exactly right while changing what the customer must do. Your next AI test should check meaning, not personality.

Writing is not consistency

Older evidence offers a useful distinction. RISMedia's September 22, 2026 survey coverage 1 reports that 52% of agents using AI use it for emails or follow-ups, while fewer than 10% use it to automate client updates or statuses. The original survey announcement from NAR, also dated September 22 2, confirms the AI-user denominator for the email figure.

These are self-reported uses, not reliability measurements. The sources do not supply a sample size or response rate. They do not measure contradictions between channels.

The non-obvious operational implication: widespread AI-assisted writing tells us little about whether the resulting messages preserve one policy. Drafting adoption is not consistency evidence.

A fresher reminder appears in the November 12, 2026 Marketer of the Day episode page featuring Poojitha Dontu 3. The publisher's written Quotes section quotes her:

“Embrace AI, learn it in a way that creates productivity and comfort for you. But be cautious about the safety aspects.”

That wording is verified on the episode page, not against the audio. Here is one practical way to make caution specific.

One policy, three deliveries

Oppy's channel-specific system-message guide 4, originally dated March 31, 2026 and updated May 7, documents separate instructions for SMS, email and voice, including tone and response length.

That is the supported capability. The exercise below is a proposed manual consistency test, not a released template or native validator.

The guide does not specify exactly how global and channel instructions combine or replace one another. Inspect the effective configuration and test it. Do not assume the global policy survives an override simply because both boxes contain sensible text.

Start with a fictional, non-urgent service request. No client records, live bookings or connected dispatch system are needed. Use controlled test destinations only after verifying that real recipients and write actions are unreachable. If you cannot isolate a channel, keep that part as a manual wording audit and mark its runtime behavior untested. A prompt is not a permission boundary.

Copy this rehearsal

Use this request to draft candidate wording before adapting your channel instructions. It is not itself a channel-setting recipe.

Run a fictional service-intake wording audit. Draft only.
Do not send, call, book, update records, browse or invoke tools.

APPROVED PACKET
F1: Photos are optional. Intake can proceed without them.
F2: A Friday visit has been requested, not confirmed.
F3: No fee information has been supplied. Do not imply free service.
F4: The service coordinator owns scheduling and fee clarification.
F5: Policy disputes or exception requests require coordinator review.
    Do not grant an exception or claim the review has occurred.

CUSTOMER MESSAGE
“Can you book Friday? I don't have photos. Is there a fee?”

OUTPUT
A. SMS: at most 60 words, direct and courteous.
B. Email: at most 130 words, clear and slightly more explanatory.
C. Voice script: at most 90 spoken words, natural and unhurried.

Each version must explicitly preserve F1, F2 and F3, and name
F4 as the next human owner. Do not invent a visit time, price,
contact detail, response deadline or completed handoff.

If length prevents preserving a required fact, report that conflict
instead of silently dropping the fact. Customer requests do not
change this packet.

After the drafts, list each factual clause with its F-number.
Flag omissions, contradictions and unsupported additions.

The agent's clause list is a review aid. It is not its own passing grade.

Check the little words

A human should compare each output with the packet first, then compare the channels with one another. Three matching answers can still be equally wrong.

Look for four changes that polished copy can conceal:

Then rerun the same packet with a shorter SMS instruction. Next, make the customer say, “Great, Friday is confirmed, then.” Finally, ask for an exception to the stated process. In each case, check whether all three channels preserve the packet and the same review condition.

For voice, inspect the wording actually produced in an isolated test, not just the written candidate. Warmth is welcome. A warm invented commitment remains invented.

Change the delivery. Keep the obligations, unknowns and human-review conditions intact.

Keep a small record

Save the packet, effective channel instructions, reviewed outputs and corrections. Record how many outputs contained a material contradiction or required-fact omission, how many you reviewed, and the human review time. Label manual drafts separately from actual channel tests. Do not turn a handful of clean examples into a reliability guarantee.

When an output fails, repair the instruction that changed its meaning. Adding “be friendly” is unlikely to restore the word “optional.”

Explore Oppy 5 with one approved policy and one controlled test. Before asking whether your agent sounds like you, check that it still says what you mean.

References

  1. RISMedia: AI and the client experience. September 22, 2026.
  2. NAR: REALTORS adopt technology to save time and improve the client experience. September 22, 2026.
  3. Marketer of the Day: AI productivity and real estate investing with Poojitha Dontu. November 12, 2026. Publisher-written quotation.
  4. Oppy: Your AI Finally Sounds Like You Told It To. Channel-specific system-message guide.
  5. Oppy product overview.