AI for Small Business study guide · Lesson 4 of 8
Marketing, Sales, and Customer Service
Create Marketing Without Losing Your Voice · Support Sales and Customer Service
This is the lesson's study guide from the FlashFeed app, plus its key facts and a self-test.
What you'll be able to do
- Use AI to adapt marketing content while preserving brand voice and substantiating claims.
- Design customer-facing assistance that uses approved knowledge and escalates unknown, sensitive, or exceptional cases.
Marketing
AI is excellent at variation. It can turn one verified source into channel-specific drafts, but it needs a brand brief: audience, promise, tone, proof points, preferred terms, prohibited claims, and examples of good work.
Build from a content atom—an original customer story, guide, demonstration, interview, or case study. Extract the core lesson and evidence, then adapt it into an email, social posts, a sales handout, or video scripts. Adaptation matters: a LinkedIn post and a customer email serve different contexts.
Keep claims truthful. AI may supply plausible performance promises, testimonials, or comparisons that the business cannot support. The owner must verify facts, permissions, links, and disclosures before publication.
Human contribution matters for intellectual property as well as quality. The U.S. Copyright Office explains that AI-assisted work can be protected when people contribute sufficient expressive authorship, while prompts alone generally do not establish authorship over a purely generated output. Use AI as part of a human creative process and preserve drafts and edits when authorship matters.
Sales Service
Customer-facing AI should be grounded in approved information: current hours, products, pricing rules, service areas, policies, and canonical answers. General model knowledge is not a substitute for your business records.
AI can help score inbound leads, but lead qualification—checking whether a prospect matches your defined needs and buying conditions—still needs a person to weigh judgment calls the data cannot capture.
Design the handoff before launch. An escalation rule tells the workflow when to stop and transfer the case. Common triggers include uncertainty, threats, legal questions, sensitive personal information, unusual exceptions, or refunds above a defined limit.
Start in draft-only mode. Have the AI propose replies while a person reviews every one. Test normal questions, incomplete questions, emotional complaints, attempts to override policy, and requests the system cannot fulfill.
Personalization should be relevant and expected. Use information the customer knowingly supplied for the relationship. Do not reveal hidden profiling or unrelated data merely because a system can access it.
Review conversation logs for recurring questions, wrong answers, failed handoffs, and policy gaps. The purpose is not to eliminate people; it is to give routine questions fast answers while preserving human attention for exceptions and relationships.
A simple brand brief
A practical brief can fit on one page. Name the primary customer, the problem you solve, the promise you can support, three proof points, the desired tone, preferred terminology, phrases to avoid, and two approved examples. Include boundaries such as “never imply same-day service” or “do not discuss a customer's account in public review responses.”
Use the brief to critique content, not merely generate it. Ask whether a draft makes an unsupported claim, sounds interchangeable with a competitor, ignores the reader's situation, or lacks a useful next step. AI is often more valuable as a tireless editor than as a source of endless new posts.
Test customer workflows safely
Create a set of test cases before enabling customer-facing automation. Include an easy question, missing information, an angry customer, a request outside policy, a privacy-sensitive question, and a prompt attempting to override the assistant's rules. Record the correct answer or handoff for each case.
Review whether the system admits uncertainty. A chatbot that says “I don't have enough information, so I'll connect you with the owner” can preserve trust. A chatbot that invents a warranty term can create a dispute.
Measure more than response speed. Track resolution, escalation accuracy, corrections, customer satisfaction, and whether staff must repair conversations later. Fast wrong answers are not an efficiency gain.
When using customer stories, reviews, photos, or messages in generated content, confirm permission and avoid exposing information that was provided for a different purpose.
Key terms
- Content atom
- A substantial source asset that can be adapted into multiple smaller channel-specific pieces.
- Escalation rule
- A defined condition that stops automation and transfers a case to human review.
Key facts
Give the Brand a Memory
A compact brand brief should define audience, promise, tone, proof points, prohibited claims, preferred terms, and examples of good work.
Create Once, Adapt Carefully
One useful customer story or guide can become an email, several social posts, a sales handout, and short video scripts when each version is adapted to its channel.
Human Authorship Still Matters
The U.S. Copyright Office says AI-assisted work can qualify for protection when humans contribute sufficient expressive authorship, but prompts alone generally do not make a purely AI-generated output copyrightable.
Ground Customer Answers
A customer assistant should answer from approved policies, product information, hours, pricing rules, and canonical answers—not from general model knowledge.
Design the Handoff
Automation is safer when it knows when to stop. Uncertainty, threats, legal questions, sensitive personal information, unusual exceptions, and refunds above a defined limit should route to a person.
Personalize with Restraint
Useful personalization reflects relevant customer context. Creepy personalization reveals data the customer did not expect you to use.
Matched pairs
Each item on the left goes with the one on the right.
- Brand voice
- The consistent personality and language choices used in communication
- Proof point
- Specific evidence supporting a marketing claim
- Lead qualification
- Checking whether a prospect matches defined needs and buying conditions
- Escalation
- Routing a case to a person with the authority or expertise to resolve it
Step by step
Turn one source into a small campaign
- Choose an original source with a clear customer lesson
- Extract the main message and supporting proof
- Draft channel-specific versions
- Check facts, claims, links, and brand voice
- Have an owner approve external content
- Measure response and record what worked
Repurposing should preserve the core truth while adapting format and emphasis.
Prepare a customer-answer workflow
- Collect the most common questions
- Write and approve canonical answers
- Identify information the assistant may access
- Define prohibited promises and escalation triggers
- Test normal, ambiguous, and adversarial questions
- Review logs and update the knowledge regularly
Start in draft-only mode before allowing automatic replies.
Self-test
Answer each question from memory, then check your work against the answer key.
1. AI drafts an advertisement claiming your service is 'guaranteed to double sales.' What should you do?
- ARemove or substantiate the claim and review it for legal and brand risk
- BReplace 'guaranteed' with 'almost certain'
- CPublish it if it sounds persuasive
- DAsk AI to add a testimonial
2. Which of the following is NOT typically part of a one-page brand brief?
- APreferred terminology and phrases to avoid
- BThree proof points that support your promise
- CA month-by-month sales revenue forecast
- DThe desired tone of voice
3. You wrote one detailed customer success story. Following the content-atom approach, what should you do with it?
- APost it only on the one channel where it performs best
- BAdapt it into an email, social posts, a sales handout, and video scripts, each tailored to its channel
- CDiscard it after a single use to keep your content fresh
- DPublish the identical text on every channel to stay consistent
4. According to the U.S. Copyright Office, when can AI-assisted work qualify for copyright protection?
- AWhenever the prompt is detailed enough to guide the output
- BNever, because AI was involved at some stage
- CAutomatically, because all AI output is protected
- DWhen a human contributes sufficient expressive authorship
5. In a brand brief, what is a 'proof point'?
- AA channel-specific version of a content atom
- BA phrase the brand has agreed never to use
- CSpecific evidence that supports a marketing claim
- DThe overall personality and language of the brand
6. A chatbot cannot find an approved answer to a customer's warranty question. What should it do?
- AEnd the conversation without explanation
- BInvent the most likely policy
- CGive a generic industry answer
- DSay it is unsure and escalate to a person
7. What is an escalation rule in a customer-service workflow?
- AA defined condition that stops automation and transfers a case to human review
- BA cap on how many messages a customer may send per day
- CA script that lets the chatbot answer harder questions on its own
- DA ranking of prospects by how likely they are to buy
8. Which of these is LEAST likely to be set as an escalation trigger?
- AA customer making a legal threat
- BA routine question about current store hours
- CA refund request above your defined limit
- DA request involving sensitive personal information
9. A customer-facing AI assistant should base its answers primarily on what?
- AThe general knowledge built into the language model
- BThe most popular answers found on competitor websites
- CWhatever reply sounds the most confident and helpful
- DThe business's approved policies, pricing, hours, and canonical answers
10. Which use of personalization in a customer reply is appropriate rather than 'creepy'?
- ABringing up data the customer never expected you to have
- BRevealing a hidden profile the system quietly built about them
- CReferencing the order the customer just placed with you
- DMentioning an unrelated purchase the customer made elsewhere
11. A customer assistant is being set up to answer product and policy questions. What should its answers be grounded in?
- AThe AI's general training knowledge
- BApproved policies, product details, hours, and pricing rules
- CCompetitor websites
- DWhatever sounds most helpful
12. Which source should a customer-facing AI treat as authoritative when answering a pricing question?
- AThe business's own approved pricing rules and records
- BThe customer's own estimate
- CA confident guess based on similar businesses
- DA general model's memorized knowledge
Go deeper
Answer key
- 1. A Remove or substantiate the claim and review it for legal and brand riskMarketing claims need truthful support. AI does not make an unsupported promise safe.
- 2. C A month-by-month sales revenue forecastA brand brief names the customer, promise, tone, proof points, preferred and prohibited terms, and examples. A revenue forecast is a financial projection, not part of the brief.
- 3. B Adapt it into an email, social posts, a sales handout, and video scripts, each tailored to its channelA content atom is one substantial source asset adapted into multiple channel-specific pieces. Adaptation matters because each channel serves a different context.
- 4. D When a human contributes sufficient expressive authorshipThe Copyright Office says human expressive authorship can make AI-assisted work protectable, but prompts alone generally do not establish authorship over a purely generated output.
- 5. C Specific evidence that supports a marketing claimA proof point is concrete evidence backing a claim. The other options describe prohibited terms, brand voice, and an adapted content atom.
- 6. D Say it is unsure and escalate to a personAn explicit uncertainty response and handoff protects trust better than a plausible invention.
- 7. A A defined condition that stops automation and transfers a case to human reviewAn escalation rule tells the workflow when to stop and hand a case to a person. Ranking prospects by how likely they are to buy describes lead scoring, not escalation.
- 8. B A routine question about current store hoursThreats, refunds above a limit, and sensitive information are common triggers to route to a person. A routine hours question is exactly what automation should answer.
- 9. D The business's approved policies, pricing, hours, and canonical answersCustomer-facing AI must be grounded in approved business records. General model knowledge is not a substitute for your own policies and pricing.
- 10. C Referencing the order the customer just placed with youGood personalization uses information the customer knowingly supplied for the relationship. The other options reveal hidden or unrelated data the customer did not expect you to use.
- 11. B Approved policies, product details, hours, and pricing rulesCustomer-facing answers should come from the business's own approved policies and records, not general model knowledge.
- 12. A The business's own approved pricing rules and recordsAn approved-knowledge design answers from the business's own current pricing rules and records, not from general or guessed information.
Sources
- U.S. Small Business Administration — AI for small business — www.sba.gov/business-guide/manage-your-business/ai-small-business
- U.S. Copyright Office — Copyright and AI, Part 2 — www.copyright.gov/ai/Copyright-and-Artificial-Intelligence-Part-2-Copyrightability-Report.pdf