How to stop free character abuse in a text to speech app
Free characters in a voice app cost cash at render time and clog the queue paying customers use. What the basic fixes catch, and where each one breaks.
You've opened the calculator on your phone, because the number on your voice provider's usage page can't be right. It says 63 million characters in nine days. All 250 of your paying customers together used 25 million last month.
The provider charges your card $500 each time you use 12.5 million characters, which normally happens twice a month. It's happened five times since the 3rd. Most of your subscriptions renew on the 1st and today's the 12th, so $2,500 has gone out almost three weeks before the next revenue comes in.
This is a worked scenario, and every price in it is an assumption you should replace with your own. Say your text to speech app gives each signup 18,000 free characters, about 20 minutes of audio at 900 characters a minute, and your premium voice costs you $40 per million characters. Nine days of text to speech free credits abuse then looks like this:
- 18,000 characters × $0.00004 = $0.72 per free account
- 3,500 accounts × 18,000 characters = 63,000,000 characters
- 63 million × $40 = $2,520
A 90,000-word novel runs to about 500,000 characters, so that's 126 novels read aloud in your best voice, at $20 a book, on your card. None of those 3,500 accounts has a payment method on file. You can ban every one of them before lunch and you still won't get a cent back, because the provider charged when the audio was made and the audio's already been downloaded.
This was the month you had planned:
- Revenue: 250 customers × $19 = $4,750
- Voice costs for those customers: 25 million × $40 = $1,000
- Hosting and everything else: $750
- Left for you: $3,000
Take off $2,520 and you've got $480 to live on.
Your inbox is full for a different reason. Each free job is 20 minutes of audio. If your workers render at four times real speed, a job holds a slot for 5 minutes, and 12 slots clear 144 jobs an hour (12 × 60 ÷ 5). On Tuesday night, free jobs arrived at 240 an hour from midnight to 6 a.m.
That's 96 more jobs an hour than you can clear, so 576 were waiting by 6 a.m., which is four hours of work (576 ÷ 144). One of your customers turns a local news briefing into audio every morning and hadn't missed a 6 a.m. upload in 14 months. Their job went to the back of that line. Their message is at the top of your inbox: "Today's edition went out with no audio. Are you down?"
Say 14 customers ask for this month back. That's 14 × $19 = $266, and your $480 is now $214. If those 14 leave as well, you lose $3,192 a year ($266 × 12). And by tonight you've got to decide what happens to the free plan, because the jobs will be back at midnight and most of those 250 customers heard a free sample before they paid.
The numbers above are invented. This next one isn't.
From our early beta
Fake users drained one team's shared model provider quota. Their paying customers couldn't use the product until the team raised its budget to cover the abuse.
That's the capacity cost of repeat claims. Free usage and paid usage draw on the same limit, so your paying customers end up waiting on traffic that was never going to pay you.
Why text to speech free credits get farmed
A character is about the easiest unit in software to collect in bulk. Text costs nothing to get hold of, it splits at any limit you set, and the audio files join back together afterward. A novel takes about 28 free accounts in this scenario (500,000 ÷ 18,000 = 27.8). On your $19 plan, with 200,000 characters a month, the same novel is 2.5 months, or $47.50.
Your product also hands over a finished good. Someone after audio needs the MP3 once, and after that your account limits and your next price change mean nothing to them. The cost lands the moment text goes in, so there's no gap between the claim and the bill where you could look at an account and think it over. And because each of the 3,500 accounts stayed under its limit, no alert fired. This is credit farming in audio form.
The obvious fix is a smaller allowance. Cut 18,000 to 5,000 and each account costs you $0.20 instead of $0.72. A novel now takes 100 accounts instead of 28, and a script doesn't mind. The person testing your voice on a real chapter gets 5.5 minutes of it and stops partway through.
The queue doesn't change either. With the same illustrative numbers:
Protect paying customers and the budget first
These put a limit on what a bad night costs, even though they say nothing about who's claiming.
Give paying customers their own render slots. Reserve 8 of your 12. Free traffic can then take 4 slots, or 48 jobs an hour, and the morning briefing renders on time however long the free line gets.
Set a daily ceiling on free characters. Say 40 real signups a day use 720,000 characters, or $28.80. A hard cap of 1.5 million characters is $60 a day, so nine bad days cost $540 instead of $2,520.
Hold rendering until the email is confirmed, so an unconfirmed signup costs you one email and no audio.
Use your cheapest voice for long free renders. Keep premium voices for a short sample. If a standard voice costs a tenth as much, 63 million characters cost $252.
The ceiling shows where all of this stops. At 48 jobs an hour, a script arriving at midnight has used up the day's 1.5 million characters by about 2 a.m., and that's only 83 full allowances (1,500,000 ÷ 18,000). Then someone signs up at breakfast to try your voice and gets told to come back tomorrow. Your cost is capped. Your free plan is also shut to real people for the day.
What you'd really like to know is whether this person has already had their 18,000 characters. The basic checks are what most teams reach for next.
The basic checks, and how each one breaks in a voice app
Start with email. Gmail ignores dots, and plus-tagged mail reaches the same inbox, so one mailbox can show up as many addresses. A unique-email rule only compares the text.
| Basic fix | What it catches | What it misses | Who it wrongly blocks | What it costs to keep working |
|---|---|---|---|---|
| Email verification | Typos and dead addresses | Any real second inbox | People whose code lands in spam | Sending and deliverability work |
| Disposable-domain blocks | Known throwaway services | New domains and ordinary mailboxes | People using privacy forwarding | A list someone updates forever |
| Gmail dots and plus tags | One inbox written several ways | A second real mailbox | People who share one inbox | Upkeep as mail services change |
| CAPTCHA | Simple scripts | People signing up by hand | Screen reader users | Fees and lost signups |
| Phone or card requirement | Casual second accounts | Anyone with another number or card | Students, and anyone who wants to hear it first | Texts to pay for, fewer signups |
| One per IP address | Repeat signups on one connection | Any other network | Schools, libraries and offices | Exceptions and tickets |
| VPN blocks | Known VPN ranges | Everything off your list | Privacy-minded people, staff on company VPNs | Another list to maintain |
| Rate limits | Bursts from one account | 3,500 accounts under the limit | Real users with a long chapter | Tuning per endpoint |
Say you block the domain behind last night's accounts. Tonight's jobs come from one that isn't on your list, and the accessibility charity that emailed you last week can't sign up because its staff use a forwarding address. CAPTCHA is a particular problem here, since people with low vision may be the most loyal users a voice app has. And at one minute per account, reviewing 3,500 accounts by hand is 58 hours, by which time all of the audio has been rendered.
Stack them up and someone who wants to hear one paragraph in your voice has to get past a confirmation code, a puzzle, a phone number and a card. Each of these rules looks at the signup, and the money goes out later, when a claim for free characters is granted.
Decide before the free characters go out
AbuseZero was built for exactly that moment, and it brings far more to it than a domain list ever could. Your one-per-IP rule ends at an address, while AbuseZero sees beyond an IP address with VPN, hosting, network reputation, Tor and proxy intelligence. With multi-layer AI risk analysis, AbuseZero examines 140,000+ signals, detects linked claims and flags abuse that single checks miss. The algorithm is proprietary, with AI that checks the claim again. Stale lists stop being your job too, because mailbox aliases, disposable domains and mail routing are all checked. AbuseZero recognizes repeat devices across new accounts, even when email addresses keep changing, and creating a fresh account isn't enough to erase previous activity. For the jobs arriving at midnight, it blocks spammy bots, browser automation and malicious scripts while allowing genuine AI agents and assistants.
You don't have to build your own abuse prevention system, or rewrite it after every bad night. Your server makes one API call before it grants the characters and gets a clear allow or deny, under rules you control. You choose how often each benefit can be claimed, so signup characters and a monthly refresh can each have their own limit, and Observe mode shows you what would be denied before anyone is blocked. Your render slots and ceilings stay where they are. The quickstart protects your first benefit.
Frequently asked questions
How many free characters should a text to speech app give away?
Enough to hear your voice on a real paragraph or two. 3,000 characters is about 3 minutes of audio, and $0.12 at the rate in this post. There's more on sizing a free allowance.
Can I take back free characters after an abuse wave?
You can remove what hasn't been used. Anything already rendered has been paid for and downloaded. See how to take back free credits.
Is a second account always abuse?
No. A narrator may keep work and personal accounts, and a household can share one connection. Write down what you mean to give each person.
Protect the free characters you give away
Start with Builder
Pay as you go. Evaluate your first claim today.
See pricing
Builder, Pro, Scale and Enterprise.
Talk to us about Enterprise
High volumes and custom terms.