This is the second tutorial in the Using Results series. In the previous tutorial you've learned how to read your feed — now let's look at one of its most useful features: the AI-generated draft reply. We'll cover where drafts come from, how to make them better, and how to turn them into replies people actually want to read.
What Is an AI Draft?
An AI draft is a suggested reply — a starting point, not a final product. For posts that score above your relevance threshold, the AI reads the conversation and writes a natural, helpful comment based on your Interest description.
It's designed to:
- Sound human and conversational, not robotic or salesy
- Reference the specific post content — not just the keyword that matched
- Offer something genuinely useful rather than pitching immediately
- Give you a head start so you're not typing from scratch every time
Every draft has the label "AI-generated draft" at the top, and you'll always see a Copy draft button to grab it for editing before you post.
Drafts only appear when you've filled in the "What to show in the draft" field in your Interest. If you left that blank, the AI won't generate a draft — it just scores relevance. This field is the single biggest lever you have over draft quality.
How Drafts Are Generated
Here's what happens behind the scenes when Engaggit finds a good match:
Reddit post (title + body)
│
▼
AI reads the full post content
│
▼
AI considers your "What I'm looking for"
and your "What to show in the draft"
│
▼
AI writes a natural reply draft
│
▼
Engaggit cleans it up on your screen
│
▼
You review, edit, and post
Two things happen after the AI writes the draft that are worth knowing about:
- The text is "humanised." Engaggit passes the raw AI output through a small post-processing step that smooths out stiff phrasing and makes the tone feel less robotic. You're seeing the polished version, not the raw AI output.
- Bold formatting is preserved. If the AI uses
**bold**in its response, it shows up as actual bold text in the draft — useful for emphasis.
What the AI Sees
To write a draft, the AI receives three pieces of information:
| Input | Why It Matters |
|---|---|
| The full Reddit post | Title and body context — the AI reads what the person actually wrote, not just the keyword that matched |
| Your "What I'm looking for" | The person's problem — the AI uses this to understand the context and write a relevant reply |
| Your "What to show in the draft" | Your specific instructions — tone, what to mention, what to offer, and how to position yourself |
The quality of a draft is almost entirely determined by how well you fill in those two description fields. A vague description gives the AI nothing to work with; a specific one gives it everything it needs.
Anatomy of a Good Draft
Here's an example of what a well-crafted AI draft looks like:
That's a frustrating situation - I've been there with complex CRMs. Quick question: are you looking for something that just handles contact management, or do you need full pipeline tracking?
If you want, I can share what we use for small teams - it's free for teams under 5. No pressure, just thought it might help.
Notice what it does well:
- Opens with empathy ("That's a frustrating situation — I've been there…")
- Asks a clarifying question rather than launching into a pitch
- Mentions something specific and helpful (free tier for small teams)
- Keeps the tone optional ("if you want," "no pressure," "just thought it might help")
- Doesn't lead with the product name — it leads with the problem
That's the kind of reply people actually upvote on Reddit.
How Your Description Changes the Draft
The "What to show in the draft" field is where the magic happens. Here's how different levels of detail change the output:
| Your Description | What the AI Writes |
|---|---|
| "Promote our CRM" | Generic, salesy, often gets downvoted |
| "Mention that our CRM integrates with Slack and is free for small teams" | More specific, still reads a bit promotional |
| "Write a helpful reply that acknowledges their frustration, asks a clarifying question, and mentions our free Slack-native CRM as one option among many — not as the only answer" | Natural, conversational, gets engagement |
The difference is night and day. A good description tells the AI how to behave, not just what to say.
What Makes a Good Description
- Tone instructions — "be casual," "sound like a helpful peer, not a salesperson," "match the subreddit's style"
- What to mention — your product, a specific feature, pricing, a free trial
- What to avoid — "don't use the product name in the first sentence," "avoid superlatives"
- Structure hints — "start with empathy, then offer one suggestion," "keep it under 3 sentences"
- Who you are — "write as a founder who uses the tool daily, not as a marketer"
The best results come when your description reads like creative direction for a writer, not a keyword list for an algorithm.
Editing Your Draft
The AI gives you a head start — but you should always read it over and make it your own before posting. Here's a quick guide:
What to Remove
| Red Flag | Why It's a Problem |
|---|---|
| Product names in the first sentence | Looks like an ad — Reddit users scroll past this |
| "Revolutionary," "best-in-class," "game-changing" | Nobody talks like this in real Reddit comments |
| Too many links | Feels promotional, even if the links are helpful |
| Generic phrases that could apply to any product | Shows you didn't read the post |
| "You should try…" | Prescriptive language puts people on the defensive |
What to Add
| Upgrade | Why It Works |
|---|---|
| A personal opening ("I had the same issue…") | Builds trust and shows shared experience |
| A specific reference to what the OP wrote | Proves you actually read their post |
| A clarifying question | Shows genuine interest, not a pitch |
| An optional next step ("happy to share more if you're curious") | Gives the OP control — no pressure |
| A conversational tone | People reply to people, not to brochures |
Before and After: A Real Example
The Reddit post:
"Been using HubSpot for 6 months and I'm drowning. It's way too much for a 4-person team. Anyone found something simpler that still has good pipeline tracking?"
Bad instructions for AI Draft:
"We built a lightweight CRM specifically for small teams - it has pipeline tracking and integrates natively with Slack so you never leave your chat. Free for teams under 5, affordable after that. Happy to walk you through it if you're interested."
After improving your prompt:
"Yeah, HubSpot is powerful but it's a lot for a small team - we had the same problem. We ended up building something lightweight that lives in Slack. Pipeline tracking, contact management, that's it - no bloat.
It's free for teams under 5 if you want to take a look, but honestly even if not, I'd recommend trying a few options with their free tiers first to see what fits your workflow."
What changed:
- Led with the shared problem ("we had the same problem") instead of the product
- Described the product in simpler, human terms ("that's it — no bloat")
- Made the pitch feel optional ("if you want," "even if not")
- Added a genuinely helpful suggestion that isn't self-promotional
When Drafts Work Best
Not every post will produce a great draft. Drafts tend to be strongest when:
- The post is detailed — more context means the AI has more to work with
- The problem is clear — "I need X because Y" gives the AI a natural structure
- Your description is specific — vague descriptions produce vague drafts
- The post is conversational — Reddit threads with personal stories get better drafts than one-liners
Posts that are very short, vague, or already have 50+ replies (where you'd be adding noise rather than value) won't benefit much from a draft — even if the AI scores them high.
What's Next
You know how drafts work and how to make them better. Now let's talk about what to do with the rest of your feed — triaging, bookmarking, and dismissing posts to stay organised.
Next tutorial → T3.3: Triaging Posts — how to bookmark, dismiss, and resolve posts, and why keeping your feed clean matters as much as finding good matches.
Frequently Asked Questions
1. Why doesn't every post in my feed have a draft?
Drafts only appear when two conditions are met: the post scored above your relevance threshold, and your Interest has a "What to show in the draft" description filled in. If that field is blank, Engaggit still scores and filters posts — it just won't generate a draft.
2. Can I regenerate a draft?
Not directly. But you can improve the result by editing your "What to show in the draft" description in the Interest settings and running a new scan — the AI will use the updated instructions for future posts.
3. What does the "humanised" step actually do?
After the AI writes a draft, Engaggit runs it through a small text-processing library that smooths out stiff phrasing, removes robotic-sounding patterns. You're seeing the final output — not the raw AI response.
4. The drafts are always too long / too short. How do I fix that?
Tell the AI what length you want. Add something like "keep it under 2 sentences" or "write a detailed 3-paragraph reply" to your "What to show in the draft" description. The AI will follow your instructions.
5. Are drafts posted automatically?
No. Drafts are suggestions only — you always review, edit, and post manually. Nothing goes to Reddit without your explicit action.
6. Can I use a different AI provider to get better drafts?
Yes. Draft quality varies between models. Cloud providers like OpenAI and Gemini generally produce more polished drafts than smaller local models. You can test different providers without changing your Interests — just switch the provider in Settings and run a new scan.
Engaggit is a privacy-first, AI-powered Reddit lead generation app. It runs on your machine, keeps your data to yourself, and helps you never miss a relevant conversation again.