Can an AI Agent Do Your Lead Generation? What It Drafts and Where You Still Approve

Quick Verdict

Yes, partly, and the part is specific. An agent can draft your posts, your service page copy and your first reply to an inquiry, then hold each one in a queue until you approve it. It will not choose your offer or pick your market, and it cannot promise leads in a market where nobody is buying.

Use it when publishing is sporadic and inquiries sit unanswered while you are on a site, including at 11pm on a Saturday. It is the wrong tool when nobody in your market is buying this quarter. Build cost runs $7,500 to $15,000 CAD per agent for a single workflow, fixed and scoped in the assessment, with complex multi-system workflows up to $30,000 CAD.

An approval-gated posting agent assembles a draft from your own material and places it in a queue with its intended publish time attached. It publishes only after a named person approves it. The gate controls permission to publish. Drafting and scheduling are separate functions.

What Canadian businesses use AI for

19.2% of Canadian businesses reported using AI to produce goods or deliver services over the 12 months before the survey, up from 12.2% a year earlier (Statistics Canada, second quarter of 2026). Among the businesses that reported using AI, the leading application was data analytics at 36.6%. Marketing automation sat at 19.8% of AI users, down from 23.1% a year earlier (Statistics Canada, Table 1).

AI use rose while marketing's share of that use fell, so the table alone does not say whether more businesses use AI for marketing. The clear part is the ranking: data analytics is the most common application, ahead of marketing automation.

Google's spam policies define scaled content abuse as generating many pages for the primary purpose of manipulating search rankings and not helping users. The policy applies "no matter how it's created", and generative AI is named as one example (Google Search Central). Neither source links the falling share to that policy, and this article does not claim a link.

Where the agent's lane ends

The table splits on who carries the consequence of a wrong call. Pricing and positioning carry consequences an agent cannot absorb. Drafting from your own material carries smaller risks: a wrong claim, a customer detail, a photo you lack permission to publish. The person who approves checks for those.

Agent or agency

Neither one guarantees leads. Results also depend on your offer, your audience and where the content is seen.

If you do not yet know what to say or who to say it to, an agency earns its retainer and an agent has nothing to produce from. If you know what to say and drafting is the bottleneck, an agent may help. If publishing is the bottleneck, clearer ownership may be enough.

The workflow, step by step

1. The agent reads from sources you already have: a photo library, a job list, a service page, a completed project folder.
2. It assembles a draft for one named slot, for example Tuesday at 9am on one channel, and attaches the publish time to the draft.
3. It runs its own checks first. Does a required detail exist? Is the claim one you have made before? Is the asset the right size for the channel?
4. The draft lands in an approval queue with the time it intends to publish.
5. A named person approves, edits or rejects. Nothing publishes on silence.
6. On approval the agent publishes through the platform API and records what went out and when.
7. The agent alerts the owner when a publish fails and records the failure. Recovery depends on the cause and on when someone can respond.

The inbound half runs on the same shape. The agent reads new submissions from your website form or a shared inbox, drafts a reply against your service list and past answers, and holds it for approval. Posts can wait for a twice-weekly queue clear. A first reply cannot, so inbound drafts push a notification the moment they are ready and you approve each one on its own. If you judge the build on time to first reply, insist on that notification path.

What the platforms actually allow

Start with eligibility. Meta's guide covers publishing "on Instagram professional accounts using the Instagram Platform" (Meta for Developers).

Then the caps. The same page gives two different numbers. Its rate-limit section says accounts "are limited to 100 API-published posts within a 24-hour moving period". Inside the carousel section the same page says accounts "are limited to 50 published posts within a 24-hour period", without saying the lower figure applies to carousels alone. Confirm the current limit for your account before the build, and do not promise a volume until you have. That page also states that filters and shopping tags are unsupported through the API and that JPEG is the only image format it accepts, so phone photos saved as HEIC or PNG need converting first.

No small business needs 50 posts a day. Reading the documentation shows what the agent cannot do, so nobody promises it to you.

Where this goes wrong

First, the approval queue is a real cost, and it is a step a build plan can underestimate. If nobody clears it on a known rhythm, drafts stack up and the channel goes quiet while the agent keeps working. The fix is a rhythm the owner actually keeps.

Second, an agent inherits whatever is in your material. A thin photo library or vague service descriptions produce thin drafts.

Third, outbound messaging to a bought list is separate work with its own rules. We scope it on its own or decline it, and it is never part of a posting agent.

What it takes to start

Five conditions, all checkable before anyone writes code.

• The channel you want is one an API can publish to, and your account on it is a professional account.
• Your source material is accurate, approved for the intended use, and sits somewhere an agent can read it.
• One person owns the approval queue and clears it on a stated rhythm.
• You pick one channel first.
• You name the number you will judge it on, and you measure it for two weeks before the build so there is a before.

For inquiry handling, measure time to first useful reply. For lead generation, also track qualified inquiries and booked work from the channel. Our lead response calculator puts a figure on the gap before you commit to anything.

The same hold applies inside a workflow, which we cover in when an AI agent answers before the work is done.

What to do next

The two-week AI Readiness Assessment maps your three highest-cost workflows in dollars, grades each one on seven factors including data availability and exception rate, and returns fixed prices rather than estimates, each holding for 90 days. Your own time is about five hours across the two weeks. It is $2,500 CAD, credited in full toward any build, plus applicable GST or HST. If lead response is one of the three, it gets graded alongside the back-office work rather than after it.