In this guide
- Stan’s take: why Firecrawl became our first choice for scraping
- The use cases and features that matter to us
- A practical play: research the reason an account might care
- Scrape, Map, Crawl, Search or Agent?
- Firecrawl pricing and a realistic cost example
- Build for incomplete pages and uncertain answers
- Firecrawl versus Apify, Discolike and Clay
- Measurement and handover
- Firecrawl buyer questions
Stan’s take: why Firecrawl became our first choice for scraping
We started using Firecrawl a couple of months ago, and it has become our first choice for web scraping. In my experience, it is the most capable scraper we have used for getting website information into an AI research workflow.
What I value is the combination of finding sources, reading them and turning the result into something an agent can use. Search matters just as much as extraction when the workflow does not already know which page contains the answer.
The expanding collector marketplace is also interesting for repeatable research jobs. For our work, the goal is not to collect an entire website for its own sake. It is to give a research agent the evidence it needs to understand a company and make the outreach more relevant.
The use cases and features that matter to us
| Use case | Feature | Why we use it |
|---|---|---|
| Find the source of an account signal | Web search | Discover relevant company pages or announcements before asking the model to interpret them. |
| Read a known company page | Scrape | Turn page content into material the research agent can work with. |
| Locate the right section of a website | Map and bounded crawl | Find relevant services, integrations or customer stories without manually opening every page. |
| Answer a structured research question | Extraction and agent capabilities | Collect the specific business facts needed for qualification or messaging. |
| Repeat a source-specific collection task | Prometheus collector marketplace | Evaluate an existing collector when it matches the research job. |
Feature references: Firecrawl Search; Example Firecrawl Prometheus marketplace collector.
A practical play: research the reason an account might care
Here is a practical way to apply that approach to your own team:
- Define a narrow question, such as whether a company serves the customer segment your offer helps.
- Search for the relevant official pages, then scrape or crawl only the useful material.
- Ask the research model to answer with supporting evidence and keep missing information explicit.
- Use the finding to qualify the account and draft a relevant message, with the source available for review.
Scrape, Map, Crawl, Search or Agent?
Pick the smallest operation that answers the question. A full-site crawl is unnecessary when the relevant service page is already known.
| Operation | Use it when | GTM example |
|---|---|---|
| Scrape | You know the page to read | Extract a product's stated target customer |
| Map | You need to discover URLs on a site | Find service, integration or case-study pages |
| Crawl | A bounded set of linked pages is necessary | Research a small relevant section of a company website |
| Search | You need to discover sources across the web | Find an official announcement or company page |
| Agent | The research needs autonomous discovery and navigation | Gather several related facts when the sources are not known |
Scrape supports readable content and structured output. For a known page, JSON mode can extract fields into a defined schema. Firecrawl's current extraction guide recommends Agent over the older Extract approach for broader autonomous research. Do not build a new workflow from an outdated tutorial without checking the current endpoint.
Firecrawl pricing and a realistic cost example
Reviewed 18 October 2026: Free provides 1,000 monthly credits. Hobby provides 5,000 credits at $19 per month on monthly billing. Annual billing has a different effective price. Larger plans increase allowances and concurrency.
The billing documentation lists one credit for a basic scraped page, one per crawled page and one per Map call. Search costs two credits per ten results, rounded up, with extra scrape charges when result pages are retrieved. JSON extraction adds four credits per page to the base scrape cost. Other options can add further charges.
For illustration, mapping 100 company sites and scraping three pages per company costs 400 base credits: 100 Map calls plus 300 pages. Requesting JSON extraction for all 300 pages adds 1,200 credits, bringing that example to 1,600. This excludes other options, retries that incur charges, additional pages and downstream tools.
That calculation explains why “one credit per page” is not the whole budget. Choose the cheapest sufficient output. Set a pay-as-you-go limit before scheduling recurring work, and monitor cost per accepted account rather than pages collected.
Build for incomplete pages and uncertain answers
A successful request does not guarantee useful evidence. You may retrieve a cookie notice, an error page, incomplete dynamic content or text from a different regional site. Validate the content before trusting the extracted fields.
- Missing content: check the retrieved page and source status before labeling the business a poor fit.
- Conflicting pages: retain both sources and request review. A newer product page may contradict an older article.
- Large websites: bound page count and relevant paths so unrelated pages do not consume the budget.
- Repeated research: store prior results and refresh based on business need, not simply because a schedule can run daily.
- Rate limits: use bounded retries and distinguish request rate from browser concurrency.
Firecrawl documents both limits. Queued jobs can consume timeout time, so increasing parallel requests may make a workload less reliable. Control concurrency and preserve job identifiers for asynchronous work.
Review a mixed sample before launch: simple sites, JavaScript-heavy sites, multilingual sites, missing pages and known wrong-fit companies. Score factual correctness and evidence relevance separately from successful HTTP responses.
Firecrawl versus Apify, Discolike and Clay
Use Firecrawl when the core job is reading and structuring web content. Evaluate Apify when a maintained source-specific Actor can collect the structured dataset you need. Evaluate Discolike when the main challenge is discovering which companies belong in a niche market. Clay can coordinate these research and enrichment steps within a broader GTM workflow.
For a simple static page, ordinary HTTP retrieval and parsing may be sufficient. The additional service is justified when it meaningfully reduces content handling, rendering or extraction work. Compare the full implementation burden, including maintenance and exception review.
These choices can complement each other. A sensible chain is account discovery, targeted website research, contact enrichment and CRM routing. Avoid using an autonomous research agent for a deterministic lookup that a smaller operation can answer.
Measurement and handover
Track useful-source rate, supported-finding rate, unknown rate, human correction rate and cost per accepted account. Then measure qualified conversations by segment. A high extraction completion rate does not establish that the research improves sales decisions.
Handover should include the qualification questions, selected page types, schema, prompts, evidence rules, refresh policy, usage alerts and recovery procedure. Document what sales should do with unknown findings. The system should remain understandable when a different person takes ownership.
frontBrick combines the commercial questions with the implementation: research design, extraction, validation and the handoff into your existing tools. Book a complimentary discovery call with Stan. Implementation pricing is custom to the scope.
Firecrawl buyer questions
Is Firecrawl an email finder?
It retrieves and structures web information. A public page may contain an address, but that is not the same as verified contact enrichment for the right buyer.
Should I crawl every page of a prospect's website?
Usually not. Start with the pages likely to answer your qualification question. Expand only when the missing information could change a commercial decision.
Does structured JSON mean the answer is correct?
No. A valid schema ensures the output has the expected shape. Evidence checks and sample review establish whether the content supports the answer.
How much does JSON extraction cost?
The reviewed billing documentation lists four additional credits per page above the basic one-credit scrape. Other selected options can increase the total further.
What is the difference between Map and Crawl?
Map discovers URLs without scraping their content. Crawl follows links and retrieves pages. Map is useful for choosing a smaller set of relevant pages before paying to process them.
Can we use it to personalize cold email?
Yes, as a research source. Use relevant, supported business facts and keep interpretation separate. Avoid turning uncertain findings into confident claims about a prospect's needs.
