AIDealsPro is reader-supported. Some links on this page are affiliate links: if you buy through them we may earn a commission at no extra cost to you. It never changes our verdicts. How we review
ElevenLabs agents are conversational AI voice agents that hold phone or web calls, billed per minute from your plan quota. They fit appointment booking, order taking, and FAQ handling — not complex judgment calls. Per-minute billing means call volume decides the plan (checked 1 Nov 2026).
Best for: Small businesses that miss calls after hours — appointment booking, order taking, and FAQ answering handled by a voice agent that never sleeps.
Not for: Situations needing real judgment — disputes, medical or legal advice, and emotional conversations that need a human.
Starting price: $0 to test (free plan, 10,000 credits/month); paid plans from $6/month (Starter), with conversations billed per minute (1 Nov 2026).
Before you start: what you need
A voice agent is only as good as its instructions. An agent given one clear job and written answers performs; an agent told to “handle calls” invents answers. Preparation is the whole project — the setup clicks take minutes.
| Have ready | Why it matters | Cost |
|---|---|---|
| One defined job for the agent | An agent without a job wastes call minutes on the wrong tasks | $0 |
| Written answers to your top 10 questions | The knowledge base decides answer quality — write it yourself | $0 |
| A plan for where calls live | Phone number, website widget, or app integration | Varies by setup |
| Test scenarios and a test number | Every test call bills minutes — plan the scenarios first | Minutes billed |
Step 1 — Give the agent one job
Pick the single highest-volume routine job: booking appointments, taking orders, answering the ten questions every caller asks. One job per agent. An agent that books appointments and also handles complaints and also sells will do all three badly.
Write the job as one sentence the agent can follow: “Book service appointments by asking for the caller’s name, phone number, and preferred day, then confirming the slot.” That sentence becomes the core of the agent’s instructions. Anything that does not fit the sentence belongs to a different agent — or to a human.
Step 2 — Write the knowledge base and guardrails
Write out the answers to the questions callers actually ask: hours, prices, location, what’s included, how to reschedule. The agent answers from what you write — it cannot know your business from your website unless you put the facts in its knowledge base.
Then write the guardrails: what the agent must never do. Never promise refunds. Never give medical, legal, or financial advice. Never quote a price you did not write down. Add escalation phrases: “Let me have someone call you back about that” beats a confident wrong answer every time.
Step 3 — Configure the voice
Pick a voice that matches the business: warm and unhurried for a clinic, crisp and quick for a delivery service. Voice choice matters more for agents than for narration — callers judge the whole business by the first five seconds. See our voice picking guide for the selection criteria.
Write the greeting word for word, including how the agent identifies itself as an AI assistant. Keep the opening short — name, business, one question. Long greetings make callers hang up before the agent has asked anything.
Step 4 — Test with real scenarios
Call the agent yourself and run the scenarios that actually happen: the caller who gives half the information, the caller who changes their mind mid-booking, the caller who asks something off-script. The off-script call is the important test — that is where the guardrails from Step 2 earn their keep.
Every test call bills minutes, so batch the scenarios into a few sessions rather than calling all day. Fix the failures by editing the knowledge base and guardrails, not by hoping the agent improves on its own. Agents get better through written instructions, not repetition.
Step 5 — Deploy and watch the minutes
Connect the agent where callers already reach you: a business phone number, a website widget, or an app integration. Start with after-hours coverage — the calls you were missing entirely — before handing it daytime traffic.
Read the call transcripts weekly for the first month. Transcripts show the questions you forgot to write answers for and the moments callers got frustrated. Add both to the knowledge base. Agents are maintained, not launched — the businesses that review transcripts get the value; the ones that set and forget get the complaints.
What agents cost per minute
Agent conversations bill per minute from your plan quota, on top of whatever the setup uses. That makes volume the whole pricing question: ten short calls a week is a different plan than a hundred calls a day. Per-minute rates are set by ElevenLabs and can change — check the current pricing page before committing.
| Volume | Example maths | Plan guidance |
|---|---|---|
| Testing (50 calls × 3 min) | 150 conversation minutes | Free plan covers testing |
| Small business (200 calls/month × 4 min) | 800 conversation minutes/month | Starter ($6/mo, 30,000 credits) starting point |
| Growing (1,000 calls/month × 4 min) | 4,000 conversation minutes/month | Creator ($22/mo, 121,000 credits) or higher |
| High volume (5,000+ calls/month) | 20,000+ conversation minutes/month | Pro/Scale — check per-minute rates first |
The honest budgeting rule: estimate your monthly call minutes, multiply by the current per-minute rate, then pick the plan whose quota covers it with headroom. See the ElevenLabs pricing breakdown for today’s plan figures and the API pricing guide for usage-based costs.
5 tips for ElevenLabs agents
- One agent, one job. Booking agents book; FAQ agents answer. Split the jobs, don’t stack them.
- Write the guardrails before the greeting. What the agent must never say matters more than how it says hello.
- Identify as AI in the first sentence. Callers who discover it mid-call lose trust in the business, not just the agent.
- Start with after-hours. The calls you were missing are pure gain — daytime handoff comes after the transcripts look clean.
- Read transcripts weekly. Every unanswered question in a transcript is a knowledge-base entry you haven’t written yet.
Common problems and fixes
The agent goes off-script
The knowledge base has a gap. Find the exact moment in the transcript, write the correct answer, and add it. Off-script behavior is almost always a missing written answer, not a model failure — the fix is writing, not settings.
Callers complain about pauses or latency
Some delay is inherent in voice agents — speech in, thinking, speech out. Shorten the agent’s responses: long answers feel slower than they are. If callers talk over the agent, tighten the turn-taking settings so it yields faster.
Costs spike in the first month
Long calls are the usual cause — an agent without guardrails lets callers chat. Cap call length, sharpen the job definition, and check the transcripts for calls that should have ended earlier. Per-minute billing punishes rambling.
Callers ask for a human immediately
Either the greeting hid that it’s an AI, or the agent fumbled the first question. Fix the greeting first — honest identification up front — then check which first questions fail. An agent that handles its one job well gets fewer human requests than an agent that pretends at five jobs.
Conversational agents: pros and cons
Pros
- Answers every call — after hours, weekends, and peak times you used to miss.
- Consistent answers from the knowledge base you wrote — no bad days.
- Scales without hiring: a hundred simultaneous calls cost minutes, not salaries.
- Full transcripts of every call — a record of what customers actually ask.
Cons
- Per-minute billing punishes long calls — costs scale with chattiness, not just volume.
- No real judgment — disputes, emergencies, and emotional calls need a human.
- Some latency is unavoidable — callers notice the pauses.
- Maintenance is ongoing — transcripts need weekly review or quality drifts.
Who should NOT use voice agents
Businesses where every call needs judgment — legal practices, medical offices, dispute-heavy services — should not hand calls to an agent. If your callers are usually upset, confused, or in a hurry, a human answers better. The main walkthrough covers the voice features that fit those businesses instead.
Frequently asked questions
What are ElevenLabs agents?
Conversational AI voice agents that hold phone or web calls: they listen, respond in a natural voice, and follow the instructions and knowledge base you write. Typical jobs are appointment booking, order taking, and FAQ handling — routine conversations with clear answers.
What do ElevenLabs agents cost per minute?
Conversations bill per minute from your plan quota at ElevenLabs’ current per-minute rate — check the pricing page, since rates can change. Budget by monthly call minutes: calls × average minutes × rate, with headroom for testing.
Which ElevenLabs plan is best for agents?
It depends on call volume, not features. Testing fits the free plan’s 10,000 credits; a small business with a few hundred calls a month starts at Starter ($6/month, 30,000 credits). High-volume operations need Creator ($22/month, 121,000 credits) or above — do the minute maths first.
Can an agent replace my receptionist?
For routine work, largely yes — booking, FAQs, and order taking are exactly what agents do well. For judgment calls, complaints, and anything emotional, no. Most businesses land on a split: the agent handles the routine volume, humans handle the rest.
What languages do agents support?
ElevenLabs voices cover 74 languages, so agents can hold conversations beyond English. Write the knowledge base in the languages your callers actually use — a translated greeting with an English-only knowledge base fails at the second question.
Do callers need to know it’s an AI?
Yes — identify the agent as an AI assistant in the first sentence. Callers who discover it mid-call lose trust, and several jurisdictions require disclosure. Honest identification also reduces frustration: callers adjust their expectations when they know.
Sources checked
Agent features and per-minute billing come from the official ElevenLabs platform and documentation; plan prices and credit amounts from the official ElevenLabs pricing page. All checked 1 Nov 2026.
Last updated 1 Nov 2026. Prices checked on the official ElevenLabs pricing page on that date; confirm at checkout.