The marketing work worth automating with AI in a small business is the repetitive layer around things you already have: content you’ve already written, leads who already contacted you, and customers who already bought. The pattern that holds up is the same every time. Automation moves the data, AI writes a draft, and a person approves anything that goes out under your name.
Below are five workflows built on that pattern, each with its trigger, its steps, the tools that fit, and the review step you shouldn’t skip. After them come the two things that decide whether AI marketing helps or hurts you: brand voice and accuracy. If terms like “trigger” and “action” are new to you, read AI Automation for Beginners first.
| Workflow | Trigger | What AI does | What a person does |
|---|---|---|---|
| Content repurposing | A new blog post or video goes live | Drafts social posts and a newsletter blurb | Checks facts and tone, approves |
| Inquiry follow-up | An inquiry with no purchase after two business days | Drafts a personal follow-up | Edits and sends |
| Social scheduling | A weekly schedule | Suggests captions for planned slots | Chooses, edits, schedules |
| Lead qualification | A new lead in the CRM | Summarizes and scores against your criteria | Decides who to contact and how |
| Review requests and replies | An order is completed; a new review appears | Drafts review replies | Approves replies, handles complaints |
Set up a content queue first
Before building any of the five, create one place where every AI draft lands. A spreadsheet or database table works fine, with columns for source, channel, draft text, status (draft, approved, rejected, published), reviewer, and notes. Automations write rows with the status “draft.” Nothing gets published until a person changes the status to “approved,” and a separate step picks up approved rows and sends them on.
This queue is your human-review step for everything that follows. It also gives you a record of what the AI produced and what you changed, which is the most useful data you’ll have for improving your prompts.
1. Repurpose each piece of content into a week of posts
- Trigger: a new item in your blog’s RSS feed, or a new row when you publish a video or podcast along with its transcript.
- Step 1: fetch the full text or transcript.
- Step 2 (AI): ask for a few short posts per social channel you actually use, one newsletter paragraph, and two pull quotes copied word for word from the source.
- Step 3: write each draft into the content queue.
- Step 4 (code): check that each pull quote appears exactly in the source text. A plain text search is enough; drop any quote that fails.
- Human step: read every draft, edit for tone, check every number, name, and link, and approve or reject.
- Tools: an RSS trigger in Zapier, Make, or n8n, an LLM step, a spreadsheet, and your scheduler.
The prompt does most of the work. A structure that holds up well:
You write social posts for [business], a [type of business]
whose customers are [audience].
Voice rules:
- [3 to 5 rules from your voice guide]
Posts we like:
[paste 3 to 5 of your best past posts]
Source:
[full article text]
Write 3 posts for LinkedIn and 3 for Instagram, based only on the source.
- Do not add facts, numbers, prices, or promises that are not in the source.
- If the source doesn't support a good post for a channel, write SKIP.
Return JSON: [{"channel": "...", "text": "..."}]The “only on the source” and SKIP instructions matter. Without them, models fill gaps with plausible claims you never made.
2. Follow up on inquiries and segment your email list
Most email segmentation doesn’t need AI. Segments built from behavior, such as “bought in the last 90 days,” “asked for a quote but didn’t buy,” “bought once,” and “hasn’t bought in a year,” are rules your email platform’s tags and automations already handle. Where AI helps is drafting one version of each sequence per segment, which you review once and save as a template. Generating fresh copy for every send adds risk without adding much value.
Inquiry follow-up is where a personal touch pays off, and AI can draft it:
- Trigger: a contact form or quote request is submitted.
- Step 1: wait two business days, then check your store or CRM for a purchase. If there is one, stop.
- Step 2 (AI): draft a short follow-up that refers to what the person actually asked about and suggests one next step, using only your product information and the form contents.
- Step 3: save it as a draft in your email account or in the content queue, not in the outbox.
- Human step: check it answers their real question and doesn’t invent availability, delivery dates, or discounts, then send it yourself.
- Tools: your form tool, an automation platform, your CRM or store, an LLM, and your email account.
Automated sequences are commercial email, and the rules apply even to small senders. In the US, the FTC’s CAN-SPAM compliance guide requires accurate headers and subject lines, a valid physical postal address, a clear way to opt out, and honoring opt-outs within 10 business days, and it makes no exception for business-to-business email. Other countries have their own rules, and many are stricter about consent, so check the law where your recipients live.
3. Batch your social posts weekly
- Trigger: a schedule, for example every Friday afternoon.
- Step 1: read next week’s planned slots from your content calendar (launches, events, themes) plus any approved repurposed drafts.
- Step 2 (AI): for each empty slot, suggest two or three caption options and alt text for the image, using the calendar notes and your voice rules.
- Step 3: add the options to the content queue.
- Human step: pick one option per slot, edit it, and schedule it in your platform’s built-in scheduler or your scheduling tool.
Keep a few things out of this workflow. Replies to comments and direct messages should come from a person, because they’re conversations, not content. Posts that react to news, anniversaries of tragedies, or anything else sensitive need a human writer. And because AI makes it easy to post more, resist that: a consistent schedule you can sustain matters more than volume.
4. Enrich and qualify new leads
- Trigger: a new lead appears in your CRM from a form, a demo request, or an event list.
- Step 1 (code): clean up the fields and check for duplicates.
- Step 2: for business leads, fetch the homepage of the company’s website.
- Step 3 (AI): summarize what the company does in two sentences and compare it against your written criteria (industry, signs of size, a location you serve, the need stated in the form). Ask for a fit rating of high, medium, low, or unknown, with a short reason for each criterion and a quote from the website or form that supports it.
- Step 4: write the summary, rating, and reasons into CRM fields clearly labeled as AI-generated.
- Step 5: notify the owner for high-fit leads, add low-fit leads to a nurture sequence, and send “unknown” to a person to look at.
- Human step: a person decides who to contact and what to say. The rating is a sorting aid, not a decision.
Two cautions. First, models guess when information is missing, especially about company size and budget, so require a supporting quote for each reason and make “unknown” the default. Second, enrichment is a privacy question. Collect only what you need to qualify a lead, be careful with third-party data about individuals, and never send automated outreach based on enriched data alone.
5. Ask for reviews and draft review replies
This workflow has two halves.
Requesting reviews:
- Trigger: an order is marked delivered or an appointment is completed.
- Step 1: wait a few days.
- Step 2: send a short message thanking the customer and linking directly to your review page.
No AI is needed here, and the rules matter more than the tooling. Ask every customer, not only the ones you expect to be happy. Google’s Maps content policy prohibits offering incentives such as discounts or free goods in exchange for reviews, discouraging negative reviews, and selectively asking only satisfied customers. That rules out the common trick of sending a “How did we do?” survey first and forwarding only the happy respondents to Google.
Replying to reviews:
- Trigger: a new review arrives, via your review platform’s notification email or integration.
- Step 1 (AI): draft a reply in your voice. For positive reviews, thank the customer and mention something specific from their review. For negative ones, acknowledge the problem, don’t argue, and invite them to continue the conversation privately.
- Step 2: put the draft in the queue. For reviews of three stars or fewer, also create a task for the owner to look into what went wrong.
- Human step: approve every reply before it’s posted. Replies are public and hard to take back, and they must not reveal private details, such as confirming what someone bought or, for health-related businesses, that they were a client at all.
Keep your brand voice
AI drafts drift toward the same polished, generic tone everyone else’s drafts have. A few inputs keep them sounding like you:
- A one-page voice guide. Three to five traits written as “this, not that,” such as “plain, not salesy” or “warm, not gushing.”
- A word list. Words you use often, and words you never use. Most businesses can ban a handful of overused marketing adjectives on day one.
- Five to ten real examples. Your best past posts and emails, pasted into every prompt. Examples steer tone more reliably than descriptions.
- A monthly read-through. Read a month of published posts in one sitting. If they blur together, update the examples and rules.
Some things are better written by you from the start: founder stories, opinions, apologies, and announcements that matter to your customers. AI can tidy the grammar afterward.
Accuracy and compliance
Models predict likely text; they don’t check facts. Google’s own guidance on generative AI content says to fact-check and review all AI-generated content for accuracy before publishing, warns that producing many pages without adding value can violate its spam policy on scaled content abuse, and notes that the same care applies to titles and meta descriptions. That advice is about websites, but it’s sound for every channel.
A short checklist for anything AI-drafted:
- Every number, price, date, and product claim matches a source you control.
- No invented testimonials, customer stories, or statistics.
- Links point where they claim to.
- No promises your operations can’t keep, such as delivery times or guarantees.
- Customer data used to personalize messages was collected with permission for that purpose.
For the data side of putting customer information through AI services, the privacy section of the small business AI automation guide covers what to send, what to strip, and which account types to use.
Measure the right things
Don’t measure AI marketing by how much content it produces. Measure each workflow against a baseline taken before you switched it on:
- Time per week spent on the task, including review.
- Edit rate: how many drafts were approved as-is, lightly edited, or rewritten. A rising rewrite rate means the prompt or examples need work.
- Outcome metrics for each channel: replies to follow-ups, click-through on newsletters, response time to new leads, and the number and average rating of reviews.
- Complaints and unsubscribes, which tell you when automation has started to feel like spam.
Where to start
Pick the one workflow closest to your biggest marketing bottleneck. For most small businesses that’s either inquiry follow-up (leads going cold) or content repurposing (good content that only gets posted once). Set up the content queue, build that single workflow, and run it in draft-only mode for a month while you track the edit rate. If you later want to add AI to your inbox, Using AI Automation to Improve Customer Support covers the same approve-before-send approach for support replies.