# The ChatGPT app and its API cite different sources.

> The ChatGPT app and its API, asked the same 5 US buying questions together, shared 13% of cited sites. 108 of 109 API citations were vendor sites.

HTML page: https://www.honeyb.ai/platform/api/chatgpt-app-vs-api

## What was measured

5 US buying questions, 3 runs of each on each side, all submitted together.

- ChatGPT app: ChatGPT app, captured in a real browser session with web search on.
- GPT API with web search: OpenAI GPT-5.6 model through an API, hosted web search on, US location.

| Question | Brands shared | Sources shared | Top pick, ChatGPT app | Top pick, GPT API with web search |
| --- | --- | --- | --- | --- |
| What accounting software should a US startup use? | 8 of 18 | 1 of 11 | QuickBooks | QuickBooks |
| What are the best billing platforms for SaaS companies? | 9 of 11 | 0 of 20 | Stripe | Stripe |
| What is the best email marketing platform for a Shopify store? | 4 of 8 | 4 of 10 | Klaviyo | Klaviyo |
| What is the best project management software for a marketing agency? | 6 of 17 | 1 of 5 | Teamwork | Teamwork |
| Which CRM is best for a 10-person sales team? | 6 of 6 | 1 of 9 | Pipedrive | Pipedrive |

Across every question the two sides shared 7 of 55 cited sources (13%) and 33 of 60 brands. 24 of the app's 37 citations were third-party sites; 108 of the API's 109 were vendor sites.

A cited page counts as a vendor site when its domain belongs to a brand named in the answers to the same question (for example quickbooks.intuit.com, docs.stripe.com); everything else is third-party (review sites, guides, publishers). Counts are cited sources across all 15 answers per surface.

AI answers change daily. That's why you need to track them. The app is what your buyers read, so that is what Honeyb captures, on every engine.

## What the API returns

Every path is on `https://app.honeyb.ai` and takes your key as a bearer token.

| Method | Path | What it does |
| --- | --- | --- |
| POST | `/api/platform/v1/tasks` | Submit a brand, up to 300 prompts and the engines to ask. |
| GET | `/api/platform/v1/tasks/{id}` | Task status, answers expected, completed and failed, and the settled cost. |
| GET | `/api/platform/v1/tasks/{id}/results` | One row per answer. |
| GET | `/api/platform/v1/tasks/{id}/citations` | One row per cited page. |
| POST | `/api/platform/v1/rank/check` | A domain and up to 1,000 keywords, checked on Google. |
| GET | `/api/platform/v1/rank/check/{id}` | Rank check status and keyword counts. |
| GET | `/api/platform/v1/rank/check/{id}/results` | One row per keyword. |

### Raw AI answers

- The answer text
- The cited sources
- Collection metadata: engine, country and time

| Field | What it holds |
| --- | --- |
| `prompt` | The question, exactly as it was asked. |
| `engine` | Which engine answered, as a stable slug: chatgpt, google_ai_overviews, google_ai_mode and the rest. |
| `answer` | The full answer text the app showed. |
| `createdAt` | When the answer was captured. |
| `url`, `domain`, `position` | Every source the answer cited, in the order it cited them, from the citations endpoint. |

### Enriched AI answers

- Everything in a raw answer
- Brands detected
- Competitors
- Sentiment
- Classified citations

| Field | What it holds |
| --- | --- |
| `brandMentions` | How many times the answer named your brand. |
| `competitorsMentioned` | Every other brand the answer named beside yours. |
| `sentiment` | The tone toward your brand: positive, neutral or negative. |
| `sentimentConfidence` | How sure that reading is, from 0 to 100. |
| `pageClassification` | What kind of page each source is: listicle, comparison, how-to, article, product page, category page, homepage or other. |

## Engines

Every engine is captured in a real browser session on the product people use, with memory and personalization off. Pass the slug in a task's `engines` list.

| Engine | Slug | Capture | Location |
| --- | --- | --- | --- |
| ChatGPT | `chatgpt` | Real browser session | Any country or US state |
| Google AI Overviews | `google_ai_overviews` | Real browser session | Any country |
| Google AI Mode | `google_ai_mode` | Real browser session | Any country |
| Gemini | `gemini` | Real browser session | Any country |
| Bing Copilot | `copilot` | Real browser session | Any country |
| Naver AI Briefing | `naver_ai_briefing` | Real browser session | Any country |
| Naver AI Mode | `naver_ai_mode` | Real browser session | Any country |
| ERNIE | `ernie` | Real browser session | Any country |
| Perplexity | `perplexity` | Real browser session | Any country |
| Claude | `claude` | Real browser session | Any country |
| Grok | `grok` | Real browser session | Any country |
| DeepSeek | `deepseek` | Real browser session | Any country |

## Prices per 1,000

US dollars per 1,000 delivered answers, from prepaid credits. No plan or subscription.

| Engine | Raw answers | Enriched answers |
| --- | --- | --- |
| ChatGPT | $1.05 | $2.05 |
| Google AI Overviews | $1.05 | $2.05 |
| Google AI Mode | $1.05 | $2.05 |
| Gemini | $1.05 | $2.05 |
| Bing Copilot | $1.05 | $2.05 |
| Naver AI Briefing | $1.05 | $2.05 |
| Naver AI Mode | $1.05 | $2.05 |
| ERNIE | $1.05 | $2.05 |
| Perplexity | $3.15 | $4.15 |
| Claude | $5.20 | $6.20 |
| Grok | $5.20 | $6.20 |
| DeepSeek | $5.20 | $6.20 |

Google rank checks: $1.00 per 1,000 checks. One check covers the top 10 results; a depth of 100 counts as ten checks.

One billable answer is one prompt on one engine, delivered. A failed capture is free, the price is held when you submit and settled to what arrived, and prepaid credits do not expire.

## Quickstart

Submit a task, poll it until its status is `completed` or `failed`, then read the answers. The key is read from the `HONEYB_API_KEY` environment variable.

```bash
curl -X POST https://app.honeyb.ai/api/platform/v1/tasks \
  -H "Authorization: Bearer $HONEYB_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "brand": { "name": "Acme Billing", "url": "https://acmebilling.com" },
    "engines": ["chatgpt"],
    "prompts": ["What is the best billing platform for SaaS?"]
  }'

# Poll until "status" is "completed" or "failed", then read the answers
curl https://app.honeyb.ai/api/platform/v1/tasks/TASK_ID/results \
  -H "Authorization: Bearer $HONEYB_API_KEY"
```

A real response: the ChatGPT answer to "best CRM for small businesses", captured in a real browser session in the United States, as `GET /tasks/{id}/results` and `/citations` return it, raw and shortened.

```json
{
  "prompt": "best CRM for small businesses",
  "engine": "chatgpt",
  "answer": "After researching the leading ...",
  "citations": [
    { "position": 1, "domain": "spotsaas.com" },
    { "position": 2, "domain": "stacknpilot.com" },
    { "position": 3, "domain": "itechguides.com" }
  ]
}
```

## MCP setup

The same key runs the MCP server at `https://platform.honeyb.ai/mcp` (Streamable HTTP), on the same credits and prices. Every config reads the key from an environment variable, so it never lands in a file, a commit or the chat.

Claude Code:

```bash
claude mcp add --transport http honeyb https://platform.honeyb.ai/mcp --scope project --header 'Authorization: Bearer ${HONEYB_API_KEY}'
```

Cursor, in `.cursor/mcp.json`:

```json
{
  "mcpServers": {
    "honeyb": {
      "url": "https://platform.honeyb.ai/mcp",
      "headers": { "Authorization": "Bearer ${env:HONEYB_API_KEY}" }
    }
  }
}
```

Codex, in `~/.codex/config.toml`:

```toml
[mcp_servers.honeyb]
url = "https://platform.honeyb.ai/mcp"
bearer_token_env_var = "HONEYB_API_KEY"
```

Tools: `list_engines_and_prices`, `create_visibility_task`, `get_task`, `get_task_results`, `get_task_citations`, `create_rank_check`, `get_rank_check`, `get_rank_check_results`. Full setup: [MCP Server](https://www.honeyb.ai/platform/api/mcp.md).

## FAQ

### Why not call the model's API yourself?

Because the app and the API cite different pages. We asked the ChatGPT app and the GPT API with web search the same 5 US buying questions at the same time: only 13% of cited sources were shared, 108 of the API's 109 citations were the vendors' own sites, and 24 of the app's 37 were third-party guides and reviews. The app is what your buyers read, so that is what Honeyb captures.

### Do AI answers change?

AI answers change daily. That's why you need to track them. The same question can name different brands and cite different pages from one day to the next, so run your prompts on a schedule and every run comes back as its own set of rows.

### Real answers or model API calls?

Real answers. Every engine, ChatGPT, Google AI Overviews, Google AI Mode, Gemini, Bing Copilot, Naver AI Briefing, Naver AI Mode, ERNIE, Perplexity, Claude, Grok and DeepSeek, is captured in a real browser session with memory and personalization off: the answer a user sees. Every row is tagged with the engine it came from.

### How is billing counted?

One billable answer is one prompt on one engine, delivered. A failed capture is free, the price is held when you submit and settled to what arrived, and prepaid credits do not expire. Raw answers start from $1.05 per 1,000, enriched answers from $2.05 per 1,000 and Google rank checks are $1.00 per 1,000 checks.

### How fast do keys arrive?

Within one working day. Apply with your work email, what you are building, the engines you need and your volume, and a person issues your key. No sales call and no subscription.

### Is there an MCP server?

Yes. The API is a remote MCP server at platform.honeyb.ai/mcp, with tools to list engines and prices, submit tasks and rank checks, and read their results. It takes the same key and spends the same credits at the same prices, and the setup prompt on this page adds it to Claude Code, Cursor or Codex.

## Get an API key

[Get your API key](https://www.honeyb.ai/platform/api/chatgpt-app-vs-api) with the form on the page. We issue keys within one working day. Tell us your engines and volume.

## Every API page as Markdown

- [Honeyb API](https://www.honeyb.ai/platform/api.md): The AI search API: the answers AI engines show people, captured in real browser sessions, read into brands and sources, with Google rank checks on the same key.
- [API Pricing](https://www.honeyb.ai/platform/api/pricing.md): Every engine's price per 1,000 answers, raw and enriched, and the price of Google rank checks. Prepaid credits, no subscription.
- [LLM Scraper API](https://www.honeyb.ai/platform/api/llm-scraper-api.md): Ask every AI engine the same prompt and get each answer with every source it cited, in one call.
- [ChatGPT Scraper API](https://www.honeyb.ai/platform/api/chatgpt-scraper-api.md): The answer the ChatGPT app gives, with every source it cited, as JSON.
- [AI Overviews API](https://www.honeyb.ai/platform/api/ai-overviews-api.md): The AI Overview Google shows above the results, with every page it cited, as JSON.
- [Google AI Mode API](https://www.honeyb.ai/platform/api/ai-mode-api.md): The answer Google AI Mode gives, with every source it cited in order, as JSON.
- [Perplexity API](https://www.honeyb.ai/platform/api/perplexity-api.md): The answer Perplexity gives real people, with every source it cited, captured in a real browser session, as JSON.
- [AI Visibility API](https://www.honeyb.ai/platform/api/ai-visibility-api.md): Every AI answer read for you: the brands named, the competitors beside yours, the sentiment and every source classified.
- [AI Mentions API](https://www.honeyb.ai/platform/api/ai-mentions-api.md): Every AI answer that named a brand, with the engine, the prompt, the sentiment, the competitors and the sources.
- [Rank Tracking API](https://www.honeyb.ai/platform/api/rank-tracking-api.md): Google position, ranking URL and whether the AI Overview cites you, per keyword, on the same key as the AI answers.
- [MCP Server](https://www.honeyb.ai/platform/api/mcp.md): Every Honeyb API call as a tool for Claude Code, Cursor and Codex, on the same key and prices.
- [Backlinks API](https://www.honeyb.ai/platform/api/backlinks-api.md): Referring domains, backlinks, anchors and new or lost links for any domain or URL. Early access.
- [Keyword Data API](https://www.honeyb.ai/platform/api/keyword-data-api.md): Search volume, CPC, difficulty, keyword ideas and Google Trends for any market, plus AI prompt volume. Early access.
- [SERP API](https://www.honeyb.ai/platform/api/serp-api.md): Where a domain ranks on Google for any keyword, the URL and title shown, every SERP feature and whether the AI Overview cites it, any country down to a city.
