AI chat for Instagram, Facebook, TikTok and SMS: a loan officer's playbook

AI chat for loan officers means one assistant, trained on your approved information, answering the messages that arrive on your website chat, text line, Facebook Messenger, Instagram DMs and TikTok (where the account’s messaging access allows). It replies quickly, answers general questions, collects details with consent and books time with you.

This playbook covers how to set it up channel by channel, the consent rules that matter, sample flows, and the mistakes that get accounts flagged. It is general information; your compliance team approves the wording you use.

Why use one AI across every channel?

Because borrowers do not care which app they used. If the answer on Instagram differs from the answer on your website, trust drops and compliance risk rises. A single approved knowledge base, a single set of rules (never quote, always hand off), and a single inbox in your CRM keep every conversation consistent and visible.

• Website chat: 34%
• SMS replies: 26%
• Instagram DMs: 22%
• Facebook Messenger: 12%
• TikTok messages: 6%

Source: Illustrative example, not measured client data

Show the data table

Example: where one loan officer's inbound messages came from in a month

Segment	Share
Website chat	34%
SMS replies	26%
Instagram DMs	22%
Facebook Messenger	12%
TikTok messages	6%

Your own mix will look different; that is the point of tracking it in a CRM rather than guessing.

How should each channel be set up?

Each channel has its own rules. The knowledge base is shared; the configuration is not.

What do good AI conversations look like?

Good conversations are short, honest and end with a clear next step. A pattern that works on every channel:

1. Acknowledge and identify. “Thanks for reaching out, I’m Jordan’s AI assistant.”
2. Answer the general question. One or two sentences, from approved content.
3. Decline anything licensed, kindly. “Rates depend on your details, so Jordan would give you a real answer.”
4. Offer one next step. A booking link, a calculator, or “reply CALL”.
5. Confirm and log. Booking confirmed, contact saved with consent, summary in the CRM.

What consent rules apply to texts and DMs?

Texts. Under the TCPA, marketing texts generally require prior express consent, and automated texts carry stricter requirements. US carriers also require business senders using standard 10-digit numbers to register their brand and campaign (A2P 10DLC). Every message flow should identify the sender, include opt-out instructions where appropriate, honour STOP immediately, and respect quiet hours.

DMs. Meta and TikTok set the rules on their platforms, including how long after a user’s last message a business may reply and what kinds of messages are allowed. The safe approach is simple: reply only to people who contacted you, keep replies relevant to what they asked, and never use DMs for unsolicited promotions.

Advertising rules still apply. A DM is still mortgage advertising. Avoid trigger terms such as specific payments or down payment percentages unless your compliance team has approved the required disclosures.

When do messages arrive, and why does speed matter?

Messages arrive when people have time: evenings, lunch breaks and weekends. That is exactly when most loan officers are offline.

• Messages received
• Messages the LO answers within 5 minutes

Source: Illustrative example of a typical pattern, not measured data

Show the data table

Example: inbound messages by hour of the day

Point	Messages received	Messages the LO answers within 5 minutes
6a	1	0
8a	3	2
10a	4	2
12p	6	2
2p	4	2
4p	5	2
6p	8	1
8p	9	0
10p	5	0

An instant, helpful first reply keeps the conversation alive until you are available, and a booked call gives both of you a time to actually talk.

How do you compare the options?

Our channels are priced individually: website chat and SMS at $69/mo each or $119/mo together, Facebook and Instagram at $59/mo, and TikTok at $59/mo; AI Engine Complete (all channels) covers all of them plus the AI Receptionist for $329/mo.

Mistakes that get accounts flagged

• Sending promotional DMs to people who never messaged you.
• Texting purchased or scraped lists, or anyone without consent.
• Sending from an unregistered number.
• Ignoring STOP, or texting at night.
• Letting the AI quote rates or promise approvals.
• Hiding that the assistant is an AI.

How do you build the shared knowledge base?

The shared knowledge base is the most important part of the set-up. Start with the questions you already answer every week, write the approved answer in two or three sentences, and add the link you would normally send. Group them by topic: programs, process, documents, timelines in general terms, contact and booking. Then write the redirects for anything that needs a licensed person.

Good knowledge-base entries are specific without being promises. “FHA loans can allow a lower down payment than many conventional loans; Jordan can confirm what applies to you” is helpful and safe. “You can buy with 3.5% down” is a trigger term and a promise in one sentence.

Review the knowledge base monthly and whenever a program, rule or process changes. Add any question the AI could not answer during the month.

What does a good hand-off look like?

The hand-off should feel like being passed to a colleague, not dropped. A good hand-off:

1. Tells the person what happens next and when (“Jordan will call you tomorrow at 2”).
2. Confirms the best contact method and records consent.
3. Passes a short summary to the loan officer so the person does not have to repeat themselves.
4. Creates a task in the CRM so the follow-up is not forgotten.
5. Lets the loan officer step into the same thread directly if they are available.

Channel-specific tips

Website chat. Open it from relevant pages, not on a timer that interrupts reading. Offer a calculator link early; many visitors want numbers before a conversation. Make the close button obvious on phones.

SMS. Keep messages short, identify yourself in the first message, and include opt-out wording. Respect quiet hours in the recipient’s time zone. Never text a list without consent.

Facebook Messenger. Reply to the question they asked, especially if they came from a specific post or ad. Keep within Meta’s response windows.

Instagram DMs. Many DMs start from a reel or story. Reference it (“Thanks for watching the first-home video”) so the reply feels connected, then offer one next step.

TikTok. Check messaging access first. Keep replies very short and link to a page that does the explaining.

How do you measure whether it is working?

Track a few simple numbers in your CRM each month rather than chasing vanity metrics:

• Conversations started by channel.
• Conversations that ended with a booked call.
• Booked calls that showed up.
• Opt-outs and complaints (these should stay very low).
• Questions the AI could not answer (to improve the knowledge base).

Compare months, not individual days, and change one thing at a time so you can tell what helped.

Privacy and data handling

Messages contain personal information, so treat them with the same care as form submissions. Store conversations in your CRM rather than scattered across apps, limit access to the people who need it, avoid collecting sensitive identifiers in chat, and describe your messaging practices in your privacy policy. If someone asks you to delete their information, make sure you can find it in one place.

The bottom line

AI chat works for loan officers when it is honest, consistent and consent-first: one approved brain, one inbox, replies only to people who reached out, and a fast path to a licensed human. Start with the channels where your borrowers already talk to you, review conversations weekly, and expand from there. For phone calls, read our guide to AI receptionists, and see each channel on the AI Engine pages.