The AI Token Markup Index

What you actually pay for AI-generated words, versus what those words cost.

Most AI content tools sell “credits” or “words per month.” Behind every credit is a token bought from OpenAI, Google, Anthropic, or DeepSeek at a public, published price. This index compares the two numbers. We update it monthly.

Last updated: June 2026. Maintained by Draft2Publish-AI. Yes, we make a BYOK plugin, so read our methodology and check our math.

The number that frames everything

A 1,500-word article is roughly 2,000 output tokens. At DeepSeek V4 Flash’s published API price ($0.28 per million output tokens), the AI cost of writing that article is about $0.0006. On Gemini 3 Flash ($3.00/M), about $0.006. On premium GPT-5.5 ($30/M), about $0.06. Every price above that is workflow, convenience, or markup. This index measures which.

Methodology (so you can check us)

1. We take each vendor’s public pricing page and the words/credits included per tier.
2. Words to tokens at 1.33x, then we DOUBLE the token budget to generously cover prompts, research calls, and retries.
3. We price those tokens at the provider’s published API rates, using the most favorable plausible model class for the vendor.
4. Markup % = (price − API cost) ÷ API cost. Every number is date-stamped.
5. Vendors can dispute any figure: corrections are published in the changelog with credit.

A worked example, step by step

Take a $49/month plan that includes 400,000 AI words. Here is the full math, with every assumption favoring the vendor:

1. Words to tokens: 400,000 words × 1.33 ≈ 532,000 output tokens.
2. Double it for prompts, research calls, and retries: ~1,064,000 tokens of billable work, call it ~1.1M.
3. Price at a cheap-but-capable model (Gemini 3 Flash, $3.00 per million output tokens): 1.1M × $3.00/M ≈ $3.30.
4. Price at a premium model (GPT-4o-class, ~$10 per million output): ≈ $11.
5. The markup: you pay $49. The tokens cost $3.30 to $11. That is roughly 350% to 1,300% over cost — and we doubled the token budget and ignored input-token discounts to be generous.

The point isn’t that $49 is unreasonable for software. It’s that almost none of the $49 is the AI. You’re paying for the wrapper, and you can’t see how much.

Camp 1: Credit resellers (markup measurable)

Sources: getgenie.ai/pricing (verified June 2026); provider API pages.

Camp 2: The opacity models (markup hidden by design)

These tools no longer let you compute what a token costs you:

• Rank Math Content AI (pricing: $4.99–$14.99/mo, annual only): migrated from per-word credits to “feature uses” — 100/500/unlimited uses per month. Its cheapest tier includes 15 long-form articles/mo: roughly $0.10–$0.30 of Flash-class tokens for $4.99. The unit changed; the economics didn’t.
• Jetpack AI (pricing: $10/mo per site): sold as “unlimited,” with a request limit that is not publicly disclosed.
• Writesonic (pricing: $16–$75/mo): credits whose cost varies per feature (a 1,500-word article consumes 5–8 credits, a caption 1–2), atop tier structures that have been renamed repeatedly.
• Elementor AI (pricing: $2.99–$15.99/mo): a shared credit pool spanning text, images, optimization, and email features, with no public credits-to-tokens table.

We don’t assign markup numbers where the vendor’s own pricing makes the calculation impossible. That impossibility is the finding.

To be clear about where we stand: a fair convenience fee is a legitimate business. Plenty of users prefer one bill and zero setup. This index measures the SIZE of the markup, not its existence.

Camp 3: The honest column (no token resale)

Credit where due — these tools charge for software, not tokens:

• AI Engine (Meow Apps) — $59/yr license, your own API keys, per-user limits you control.
• AI Puffer (formerly AI Power) — free core, BYOK across providers, “no hidden credits” is their own line.
• Divi AI — flat subscription, apparently genuinely unlimited (we’ll verify the fine print and update).
• Official WordPress “AI” plugin — free, BYOK via the WP 7.0 Connectors screen.
• Copy.ai — genuinely free tier with unlimited words; paid plans price the platform, not tokens. (Listed honestly: free-for-real belongs here.)
• Draft2Publish-AI — that’s us: free on WordPress.org, BYOK, zero token markup; paid tiers price the workflow, never the tokens.

Why this matters now

WordPress 7.0 put API key management into core. The platform itself has decided users should own their AI relationship. The credit-resale era had a good run; this page is its scoreboard on the way out.

Common questions

How much does AI content actually cost per article?

At published June 2026 API rates, a 1,500-word article is about $0.0006 of output tokens on DeepSeek V4 Flash, about $0.006 on Gemini 3 Flash, and about $0.06 on GPT-5.5. Even doubling those budgets for prompts, research, and retries, the raw AI cost of an article is cents at most.

What is the markup on AI writing tool credits?

It varies by vendor and model class. In the clearest measurable case in this index, a $49/month 400K-word plan implies roughly 400% to 1,300% over the published API cost of those words, depending on which model class actually serves them.

Is BYOK cheaper than an AI content subscription?

For the AI usage itself, almost always: with your own API key you pay the provider’s published rate with no intermediary. A subscription can still be worth it for the workflow and software around the AI. The honest comparison is software value versus software price, with tokens at cost either way.

Why can’t I calculate what some tools’ credits cost?

Because several vendors now price in units that don’t map to tokens: “feature uses,” undisclosed request limits, or credit pools shared across unrelated features. When the unit can’t be converted, the markup can’t be computed. We list those tools separately rather than guessing.

Changelog

• June 2026: first edition. 8 vendors examined.

Disputes

Vendor and spotted an error? Email [email protected]; corrections published here with credit and date.