AI Foundations for Small Business · Lesson 2 of 11

The AI Tool Landscape

Time: ~25 min reading + ~20 min worksheet (live exploration) Goal: Know the major players, their strengths, and how to pick the right one for a task. By the end, you'll have opinions.

---

In 2026, there are four general-purpose AI assistants that matter most for SMB work, plus one open-source wildcard. You should have an account on at least two of them before finishing this module.

1. ChatGPT (OpenAI)

• Who runs it: OpenAI
• Where to get it: chatgpt.com
• Consumer pricing (as of this writing — check the site): Free tier; Plus around $20/mo; Team and Enterprise tiers for businesses
• Strengths: Most recognized brand, biggest ecosystem of integrations and custom GPTs, solid image generation built in (DALL-E), strong voice mode
• Weaknesses: Historically more prone to hallucination than Claude on careful writing tasks, privacy guarantees on the consumer product are weaker than business-tier offerings
• Use it when: You need an all-rounder, you want a massive library of pre-built custom GPTs, or your client is already paying for it

2. Claude (Anthropic)

• Who runs it: Anthropic
• Where to get it: claude.ai
• Pricing: Free tier; Pro around $20/mo; Team and Enterprise tiers
• Strengths: Generally the best in class for long-form writing, careful reasoning, and code. Large context windows (you can dump entire books or codebases in). Claude Code for developers. Projects feature for persistent context.
• Weaknesses: Smaller ecosystem than OpenAI, somewhat slower on simple tasks, more careful/hedging on some topics
• Use it when: You're drafting anything long, analyzing big documents, writing code, or need thoughtful back-and-forth on strategy

3. Microsoft 365 Copilot

• Who runs it: Microsoft (using OpenAI models under the hood, plus their own)
• Where to get it: Built into Word, Excel, PowerPoint, Outlook, Teams. Requires an add-on license per user.
• Pricing: Roughly $30/user/month on top of a qualifying M365 license. Check current pricing — it moves.
• Strengths: Lives *inside* the tools your clients already use. Can reference the user's own emails, files, and Teams chats. Enormous productivity unlock if a company is already deep in M365.
• Weaknesses: Requires M365 licensing first. Quality has been uneven — great at some tasks, underwhelming at others. Admin setup matters a lot.
• Use it when: Your client lives in Outlook/Excel/Teams all day and the friction of switching to a separate chat app is the blocker

4. Gemini (Google)

• Who runs it: Google
• Where to get it: gemini.google.com, plus embedded in Workspace (Docs, Gmail, Sheets)
• Pricing: Free tier; paid tiers bundled with Google One / Workspace add-ons
• Strengths: Deep integration with Google Workspace (Gmail, Drive, Docs, Sheets). Best-in-class for tasks involving Google's own data (YouTube, Maps, Search). Very large context windows.
• Weaknesses: Product brand is still muddled; Google has changed names and tiers repeatedly. Quality is strong but has historically been less consistent on pure writing tasks than Claude.
• Use it when: Your client is a Google Workspace shop, or when you need deep search/research

The Wildcard: Local / Open-Source Models

• Who runs them: Meta (Llama), Mistral, Alibaba (Qwen), and others — you download and run them yourself
• Where to get them: Ollama (ollama.com), LM Studio, Hugging Face
• Pricing: Free to use; you pay for the hardware
• Strengths: Data never leaves your machine (huge for healthcare, legal, any regulated industry). No per-token costs. Improving fast.
• Weaknesses: Requires a decent machine (a Mac with Apple Silicon or a PC with a good GPU). Models are still a step behind frontier commercial models on raw smarts. Setup is a project.
• Use it when: Data privacy is non-negotiable, or volume is so high that API costs would hurt

Worth knowing: Local LLMs are a great story for healthcare practices, law firms, accounting firms, and anyone with HIPAA, PII, or regulated data concerns. You can honestly say "nothing leaves your building." That's a very different conversation than "we trust OpenAI's data policy."

This distinction confuses most beginners. There are two ways to use every one of these models:

Consumer Chat Apps (chatgpt.com, claude.ai, etc.)

• You log in, type in a box, get replies
• Priced per user per month, flat fee
• Comes with a UI, chat history, and conveniences
• What you'll use 80% of the time

API (Application Programming Interface)

• Developers call the model from their own software
• Priced per token (fractions of a penny per query, but it adds up)
• No UI — you build your own
• How custom AI products get built (including everything AAT builds for clients)

Why you care as an opportunity spotter: "We want to add a chatbot to our website" is an API project. "We want our CSRs to write better emails" is a consumer-app rollout. The difference in effort and cost is huge.

You don't need accounts on all of these, but you should recognize the names:

| Tool | Category | What it does | |---|---|---| | Perplexity | Research / Search | AI answers with live web search and citations | | Notion AI | Embedded in Notion | Writing help inside your Notion docs | | Grammarly | Writing assist | Still incredibly useful; AI-powered rewriting | | Jasper, Copy.ai | Marketing content | Pre-templated content generation | | Otter.ai, Fireflies | Meeting transcription | Records + summarizes meetings | | Synthesia, HeyGen | AI video | Generate video avatars from scripts | | ElevenLabs | Voice | Clone voices, generate narration | | Midjourney, DALL-E, Stable Diffusion | Image generation | Create images from text prompts | | GitHub Copilot, Cursor, Windsurf | Coding | AI pair-programmer for developers | | Zapier, Make, n8n, Power Automate | Automation | Connect AI to other apps, trigger workflows |

Most of these are built on top of the big four models. They add UI, templates, and workflow on top.

For most SMB conversations, here's the decision tree:

• Is the client deep in Microsoft 365?

→ Start with Copilot. They're already paying for the platform.

• Is the client deep in Google Workspace?

→ Start with Gemini in Workspace.

• Neither, or looking for "the best"?

→ Claude for writing, analysis, and long documents. ChatGPT for everything else.

• Regulated data, on-prem requirements, or massive volume?

→ Consider local/open-source models. Budget for setup and hardware.

• Custom chatbot, automation, or app?

→ That's an API project. Big four + a developer.

This will change. Don't memorize it — understand the *logic* (platform fit, privacy, use case) and re-run it as the tools evolve.

The biggest mistake beginners make is recommending a tool they haven't used on the exact task the client cares about. The big four are pretty close on most things, but they have real personality differences, and the right one often depends on the specific job.

Rule: Before you recommend a tool for a client, run the client's actual task through at least two of the big four yourself. You'll be shocked how often the "obvious" choice isn't the best one.

---

• Big four: ChatGPT, Claude, Copilot, Gemini. Know the strengths of each.
• Consumer chat apps (per-seat) vs. APIs (per-token, for custom builds) — different worlds, different costs.
• Platform fit (M365 vs. Workspace) often beats raw model quality for SMB choice.
• Local models are a real option for privacy-sensitive clients.
• Always test on the actual task before recommending. Your own direct experience is your differentiator.

---

Next up: Module 3 — How to Prompt. The part of the job that most people do badly.

---

Time: ~20 min (mostly hands-on) Goal: Get accounts, run the same task across tools, form your own opinions.

---

If you don't already have them, sign up for free accounts on at least two of the following:

• [ ] ChatGPT (chatgpt.com)
• [ ] Claude (claude.ai)
• [ ] Gemini (gemini.google.com)
• [ ] Microsoft Copilot (copilot.microsoft.com — the consumer version is free)

Bonus: if you have a machine with 16+ GB of RAM, install Ollama (ollama.com) and pull a small model like `llama3.2` to see what local inference feels like.

Run this exact prompt on three different tools (pick any three of the big four):

> "You're helping a 25-person HVAC company in the Northeast. Write a 150-word email to their existing customer base announcing a new preventive maintenance membership program at $19/month. Tone: friendly, practical, a little folksy. No cheesy sales-speak."

Record:

| Tool | Did it follow length? | Did it hit tone? | Anything weird or notable? | Overall grade A–F | |---|---|---|---|---| | | | | | | | | | | | | | | | | | |

Which one won? Why? _____________________________________________ _____________________________________________

For each client, pick which tool you'd start with and justify in one sentence.

• 28-person accounting firm. Everyone lives in Outlook and Excel all day. Existing M365 Business Premium.

→ Tool: ______________ → Why: ______________

• Solo attorney in a small practice. Handles sensitive family law cases. Owns a new MacBook with M-series chip. Paranoid about cloud data.

→ Tool: ______________ → Why: ______________

• 60-person marketing agency. Google Workspace shop. Clients ask them to produce a lot of written content — blog posts, captions, email campaigns.

→ Tool: ______________ → Why: ______________

• 15-person equipment dealership. Doesn't use cloud productivity apps at all — everything is done in Outlook desktop and Word docs. Wants "AI" but has no other stack requirements.

→ Tool: ______________ → Why: ______________

• Manufacturing company. Wants to build a chatbot trained on their own 10,000+ machine manuals and service records. Needs it on their internal portal.

→ Tool/approach: ______________ → Why: ______________

---

Exercise 2: There's no single right answer — the point is to build your own reference experience. Most people find Claude gives the strongest copy on this kind of warm-tone writing task, ChatGPT is very close, and Copilot/Gemini are solid but sometimes a little more generic. Your mileage will vary.

Exercise 3:

• Copilot — they're already paying for M365, so Copilot puts AI inside the tools they already use. No new app to learn.

• Local model via Ollama/LM Studio — the M-series MacBook can easily run a 7B–13B model locally. Nothing leaves the machine. Privacy-first positioning is exactly the story a family law attorney needs.

• Gemini in Workspace (or Claude/ChatGPT for the content itself) — Workspace integration is the platform fit, but the agency will likely use multiple tools. A realistic answer: Gemini for Workspace-embedded work, Claude for the highest-quality long-form writing.

• Claude or ChatGPT (consumer Pro tier) — no platform fit to optimize for. Pick based on task quality. Both are great starting points.

• API + RAG build. This is not a consumer chat question. This is a custom project: you'd index their manuals and service records into a vector store, put a RAG pipeline in front, and build (or buy) a chat UI for their portal. The model choice for the backend could be any of the big four via API.

---

Check yourself: Could you walk into each of those five conversations and defend your recommendation with one clean sentence? If yes, you're ready for Module 3.

Previous: What AI Actually Is (and Isn't) · Next: How to Prompt