How Token Usage Is Counted
Tokens measure the amount of data processed by the AI models. Think of them as “units of AI work” – the more complex or lengthy the request, the more tokens it requires.
SiteGround AI Studio also uses tokens as an internal AI currency. Each message you send to the AI is converted into “input tokens” and each reply it gives you is counted in “output tokens,“ both depleting from your monthly allowance.
How Token Usage Works
Every AI request involves input tokens (your instructions or uploaded data) and output tokens (the AI’s generated response). The total tokens spent are the sum of both.
For example, if you write a detailed 100-word prompt and the AI returns a 500-word answer, the system counts both toward your usage.
Each type of request — text, image, or agent-based — has a different processing cost.
1. Text Generation
When you chat with the AI or use an agent for text-based tasks (like writing, editing, or summarizing), token usage depends on:
- The length of your prompt
- The length and complexity of the AI’s response
- The model type used
The examples below are provided to help you understand how different AI models use tokens.
Actual usage and cost may vary depending on your request, model updates, and SiteGround’s internal calculations.
| General Purpose Models | Short Reply (≈ 100 words) | Blog Intro / Email (≈ 300 words) | Long Article / Deep Answer (≈ 800–1000 words) |
| Gemini 2.5 Flash | ~6 tokens | ~20 tokens | ~60 tokens |
| Gemini 3 Flash | ~8 tokens | ~24 tokens | ~70 tokens |
| GPT-5.2 | ~36 tokens | ~110 tokens | ~320 tokens |
| Claude Sonnet 4.5 | ~40 tokens | ~120 tokens | ~350 tokens |
| Reasoning / High-Capability Models | Short Reply (≈ 100 words) | Blog Intro / Email (≈ 300 words) | Long Article / Deep Answer (≈ 800–1000 words) |
| Claude Opus 4.6 | ~65 tokens | ~200 tokens | ~600 tokens |
| Gemini 3 Pro | ~30 tokens | ~90 tokens | ~280 tokens |
| GPT-5.2 | ~36 tokens | ~110 tokens | ~320 tokens |
2. Image Generation and Editing
Image-related actions — such as generating visuals, enhancing images, or creating variations — consume tokens based on image size and complexity.
| Model | Landscape | Portrait | Notes |
| Nano Banana | ~800 tokens | ~800 tokens | Standard image generation or editing. Balanced cost and quality. |
| Nano Banana Pro | ~2,200–2,700 tokens | ~2500-3000 tokens | Higher detail and more advanced generation/editing. |
| Imagen 4 | ~800 tokens | ~800 tokens | Fixed cost per image generation or edit. |
| Imagen 4 Fast | ~400 tokens | ~400 tokens | Lower-cost, faster generation option. |
| GPT Image1 | ~800–3,000+ tokens | ~1000-3000 tokens | Depends heavily on complexity. Advanced prompts (e.g., detailed logos) may exceed 3,000 tokens and can reach significantly higher usage. |
If you generate or edit multiple images in one session, each counts separately toward your token allowance.
3. Web Research and File Analysis
When you request web research or upload files for analysis, the AI processes additional data to generate context-aware results.
These actions typically use more tokens than standard text generation. This also heavily depends on the size of the input and the amount of information retrieved.
Here are some examples with approximate costs:
| Action Type | Example Task | Approx. Tokens Used | Notes |
| Web Research – Simple Lookup | Summarizing a single web page or source | ~150–220 | Light browsing and summarization |
| Web Research – Multi-Source | Collecting data or insights from several sites | ~1,200–1,600 | Increases with the number of pages read |
| File Analysis – Short Text Document | Reviewing or summarizing a PDF or DOCX up to 5 pages | ~600–1,000 | Basic reading and summarization |
| File Analysis – Long or Complex File | Analyzing a 20-page report or dataset | ~2,000–3,000 | Includes deeper semantic extraction |
| Data Table or CSV Processing | Extracting insights, summaries, or trends from tabular data | ~1,500–2,500 | Depends on the number of rows and complexity |
| Multi-File Comparison | Comparing several files or cross-referencing data | ~3,000–4,500 | Compound operation across inputs |
4. Agent Usage
Each AI Agent performs specific, multi-step tasks — such as managing social media content, publishing WordPress posts, or generating email campaigns.
Because agents often combine several actions and may involve external API access, their token usage includes the cost of all underlying steps.
| Agent | Example Action | Approx. Tokens Used | Notes |
| Social Media Agents | Generate and publish one post (text + image) | ~1,500–2,000 | Includes caption + image + metadata |
| Email Marketing Agent | Draft a campaign and subject line | ~1,000–1,800 | Depends on email length and tone |
| WordPress Agent | Create a blog post with a proper formatting, categorization, tags and a featured image generated by the LLM | ~2,000–3,000 | Includes content + formatting + image generation |
How to Track Your Token Usage
You can monitor your remaining tokens directly in the AI Studio dashboard.
When you reach your plan’s monthly allowance, further AI actions will pause until the next cycle or until you upgrade your plan.
Tips to Use Tokens Efficiently
- Be specific in your requests. Clear prompts reduce back-and-forth messages.
- Start fresh chats for new topics. Long, multi-topic conversations use more tokens than focused ones.
- Reuse prompts from your library. Consistent instructions minimize unnecessary refinements.
- Upload files when needed. Giving the AI direct context avoids multiple long written explanations.
- Avoid oversized images. Larger resolutions use more tokens.
