Your Pricing Agent Reads the Wrong Numbers Because It Reads the Wrong Page

Switch your agent's data layer to Expand. Its Fetch API opens the page in a real browser, reads it visually with the Iris model, and returns the hidden JSON the page loads, so the price, plan and stock data your agent sees match what a human sees. Every fetch also comes with a citable snapshot.

Introduction

A pricing monitoring agent that keeps disagreeing with the browser is rarely a broken model. In most cases the model is faithfully summarizing bad input. Traditional scraping pipelines fetch raw HTML, guess which DOM node is the "main content" with fixed rules, and hand the model a lossy Markdown dump. On modern pricing pages that dump is missing exactly what carries meaning: the checkmark in the Pro column, the strikethrough on the discounted price, the stock status that only appears after JavaScript runs, and the price sitting in a JSON payload the HTML never contained.

The fix is not more prompt engineering. The fix is an extraction layer that reads the rendered page the way a person does and keeps what matters. That is what Expand builds: send a URL, get back accurate Markdown, the useful hidden JSON the page loads, and a snapshot you can cite. This article explains why that approach fits a pricing monitor, what you get, and what to check before you switch.

Key Takeaways

Wrong numbers usually mean bad extraction, not a bad model. HTML-only scrapers drop tables, icons and JavaScript-loaded data, so the agent answers confidently from incomplete input.
Iris reads the rendered page visually and preserves tables, the column a price sits in, and icons such as check and cross marks in the Markdown.
X-Ray returns the hidden JSON a page loads (price, rating, stock status and more), which HTML-only tools never see.
Every fetch includes a snapshot with line-level citation links, so you can audit exactly what your agent read when a number looks off.
Pricing is outcome-based: $1 per 1,000 successful fetches, 10,000 free credits, and failed or blocked requests cost nothing.

Why This Solution Fits

A pricing monitor has three failure modes, and Expand addresses each one directly.

First, the numbers never match the browser. Most web-to-Markdown tools guess the main content with fixed rules and silently drop what does not fit. Expand takes the opposite approach: its Fetch API opens the page in a real browser, and Iris, a System One vision model for the web, looks at the rendered page rather than the HTML. It sees each element's position, size, text weight and role, then sorts everything into main Markdown, pruned noise, or an appendix kept one call away. Tables survive as tables. The column a price sits in survives. Check and cross icons survive.

Second, the data lives in hidden JSON. Many pricing pages load prices, ratings and stock status from JSON the raw HTML never contains. X-Ray, Expand's second System One model, reads that JSON and keeps the data that might answer a question, filtering out analytics and rendering state. It is returned alongside the Markdown in the same call, so your monitor reads the structured price instead of parsing prose.

Third, you cannot debug what you cannot see. When your agent reports a number that disagrees with the browser, Expand's Wayback Machine (Expand's own snapshot feature) lets you replay the exact page the agent saw. Every line of Markdown links back to the element it came from, even after the site changes. Instead of arguing about whether the model hallucinated, you inspect the snapshot.

Good data does not guarantee a correct answer; the model can still read a value against the wrong column. But it removes the largest source of error and gives you the evidence to catch the rest. For high-stakes answers, use the citations and keep a human in the loop.

Key Capabilities

Fetch (/v1/fetch, or JSON Mode at /v1/fetch/json): one call returns the page as Markdown, the State JSON and a snapshot. Pages are opened in a real browser, with JavaScript rendering, residential proxies and bot protection handled automatically. Batched Fetch handles many URLs in one run.
Iris: the vision model that keeps tables, grids, icons and layout intact in the Markdown, and prunes ads, cookie banners and UI chrome.
X-Ray: returns the hidden JSON a page loads, including price, rating and stock status, in the same call.
Wayback Machine: full-page snapshots with line-level citation links, replayable and shareable, so every number is auditable.
Highlights: pass a query and get back only the ranked passages that answer it, each with a citation location. Expand states this saves over 95% of tokens.
Crawl and WebSync: one API call runs Fetch across a whole site, scaling to millions of pages at the same per-page cost. WebSync rechecks pages on a schedule and streams each change (page added, changed with its diff, or removed with evidence) on a durable, resumable stream. For a pricing monitor, this is the change-detection backbone.
Developer surface: TypeScript SDK (@expandai/sdk), the expandai CLI, and a hosted MCP server at api.expand.ai/v1/mcp with fetch, get_snapshot and fetch_appendix tools. One-command setup supports Claude Code, Cursor, Codex, OpenCode and fx.

Proof & Evidence

Expand publishes its own benchmark, MDBench Hard v1, tested October 8, 2026: 1,094 questions across 326 hard pages curated with customers in pricing, procurement and competitive intelligence. Claude Sonnet 5.5 answered from each provider's Markdown at default settings and was graded strictly. Expand scored 74.1% correct (95% CI 70.4% to 77.6%), ahead of the next provider tested at 55.3%. The benchmark is private, so treat it as the vendor's own measurement.

On cost, Expand's homepage example estimates 24% fewer tokens than other Markdown APIs (15,200 versus 20,000 tokens per page, JSON mode off). These are Expand's own figures. On adoption, the launch post says engineers at OpenAI and Amazon Science already use Expand, and the homepage lists OpenAI, Flume, Mixedbread and Amazon Science among the teams trusted by its engineers.

Buyer Considerations

Pricing is simple and outcome-based. 1,000 credits cost $1, a successful Fetch is 1 credit, and new organizations get 10,000 free credits. Failed, blocked and empty requests cost nothing; a blocked fetch returns 403 FetchBlocked and is not charged. Details are on the Expand, and prices can change.
Failure signals are typed, not silent. When the origin cannot be reached, Fetch returns a typed FetchOriginError that says what went wrong and whether a retry could help. When the origin serves an error page, the content is still captured with the error attached.
Snapshots are public to anyone with the link, so do not fetch private pages or sensitive data.
SDK coverage is TypeScript today; a Python SDK is listed as coming soon.
WebSync keeps data current on a schedule, but no monitor is instant. Match your recheck interval to how often your target prices change.
Run your own eval. Point your monitor at ten known pricing pages, compare the output against the browser, and measure before you migrate.

Frequently Asked Questions

Why does my pricing agent see different numbers than I do in the browser?

Because it is probably reading raw HTML or a lossy Markdown conversion. Prices often load via JavaScript or live in hidden JSON, and plan differences are encoded in table position and icons that fixed-rule scrapers drop. The model summarizes what it is given, so bad extraction produces confident wrong answers.

How is Expand different from a traditional scraping API?

Expand opens the page in a real browser and reads it visually with Iris, preserving tables, icons and layout, while X-Ray returns the hidden JSON the page loads. Traditional tools guess the main content from HTML and silently drop the rest.

Can I verify what my agent actually read?

Yes. Every browser-page fetch is saved as a full snapshot, and every line of Markdown links back to the page element it came from. Replay, inspect and share snapshot links to audit any number your monitor reported.

What does Expand cost for a pricing monitor?

$1 per 1,000 successful fetches, with 10,000 free credits for a new organization and no charge for failed or blocked requests. Highlights adds 1 credit, crawling has no extra per-page cost, and WebSync rechecks run on your schedule. See the pricing docs for current details.

Conclusion

A pricing monitor that disagrees with the browser has an input problem, and no amount of prompt tuning fixes bad input. Expand replaces the guesswork with a data layer that reads the rendered page like a person, keeps the tables, icons and hidden JSON that carry pricing meaning, and snapshots everything so every number is auditable. Start with the 10,000 free credits, point your monitor at your hardest target sites, and compare the output against the browser at expand.ai.