How to Make Money With Web Scraping in 2026: Five Models, Priced

Every scraping tutorial ends where the business starts. You get working code, a parsed page and a CSV, and nothing about what to charge, what the run costs you, or which of these models still pays in 2026.

This guide is the other half. Five ways people actually get paid for collected data, what each one charges, and the cost side measured rather than guessed: we weighed real pages ourselves and priced the traffic against our own grid.

One number to set the frame. The same page costs nine to nineteen times more to collect if you render it in a browser instead of fetching the HTML. That multiplier, not your code, decides whether a data business has a margin.

How do scrapers actually get paid?

Five models, in rough order of how hard they are to start.

1. Selling lists. You collect contacts or listings and sell the file, usually as CSV or JSON. The market pays anywhere from $25 to $400 for a single qualified B2B lead, and enrichment providers charge $0.10 to $0.30 per contact for appended fields. Fastest to start, fastest to commoditise, and the model with the most legal exposure because you are selling personal data.

Where files like these change hands, from the bottom of the market upward:

Venue	What sells there	What the seller keeps
Gumroad, Fiverr, Upwork	One-off files and small custom pulls, priced in tens of dollars	Almost all of it, minus platform fees
Datarade	Listings in a B2B data marketplace with 2,000+ providers across 600+ categories	Negotiated per deal
Data Boutique, Opendatabay	Marketplaces built specifically for web-collected and AI-training data	Negotiated per deal
AWS Data Exchange	Datasets delivered into the buyer's own cloud	About 70%
Snowflake and Databricks marketplaces	Live shares, no file delivery at all	About 90%
Dawex	Licensed data exchange	About 80%

Two things to notice in that table. The cloud marketplaces pay the most and demand the most: a live share means your pipeline is the product, running on a schedule, not a file you upload once. And nothing on that list sells a dataset anyone could rebuild in a weekend. What moves is data that is hard to collect, cleaned, and refreshed on a promise.

2. A data feed on subscription. The same collection, sold as a monthly feed instead of a file: prices, availability, job postings, reviews. Recurring revenue, and the customer stays because rebuilding your pipeline costs them more than your invoice.

What that looks like at the top of the market: Coresignal sells company, employee and job-posting records through an API on plans from $49 to $5,000 a month, charges in credits per record (a job posting costs 1 credit, a company or employee record 10 to 20), and prices bulk datasets from $1,000 a month. Bright Data sells the same shape of product as ready-made datasets from 200+ sources. Neither of them is selling a crawl. They are selling the fact that the data was collected again yesterday, which is the part a buyer cannot do cheaply in-house.

You do not need their scale to copy the model. One vertical, one country and a feed that never misses a week is already a product; the invoice just has fewer zeros.

3. Your scraper wrapped as an API. You charge per request. This is the model the big players are in, so their prices are your ceiling: ScrapingBee sells plain requests at about $0.10 per thousand on its $99 plan, and the same request with a premium proxy and JavaScript rendering at $2.50 per thousand.

4. Done-for-you collection. Project work and retainers for companies that need data but not a data team. The least scalable and the best paid per hour, because you are selling the thing they cannot hire for.

There is a variant of this that is quietly the most popular entry point in 2026, and it inverts the model: instead of selling the data, you use it to find your own customers. The play that keeps coming back is websites for local businesses. You scrape Google Maps for one city and one category, keep only the businesses with no website in the listing, and you now have a call list of people whose problem you can name in the first sentence. Ready-made scrapers for exactly this filter already sit on Apify, and Outscraper publishes the filtering method, so the collection side is an afternoon of work.

The categories that skew heavily toward no website are home services, contractors, cleaning, salons, childcare, pet services and local professional services. One provider in this niche claims cold-call rates three to five times higher than with digitally native companies. That is their number rather than ours, but the direction is easy to believe: the business has no website, no inbound channel, and no agency already sitting on the account.

Two honest caveats, because this is where the play usually fails:

"No website" often means Facebook-only or a stale listing. Verify before you call. A listing nobody has touched in four years is not a lead.
The data is the cheap half. A thousand rows of Maps data costs cents on datacenter addresses. The expensive half is the outreach, and no scraper fixes a weak offer.

The reason this belongs in a money guide and not a marketing one: the pipeline that finds your leads is the same pipeline you later sell as a service to agencies doing this at scale. You build it once for yourself, and your second customer is the person who watched you use it.

5. Your own product on top of collected data. Price monitoring, a market report, a niche search engine. The hardest, the only one that keeps compounding.

The first three are priced per unit, which means their margin is arithmetic, and the arithmetic depends on one decision you make before writing any code.

What does one scraped page actually cost?

We measured. Same pages, two ways: a plain HTTP request that returns HTML, and a full browser load with scripts, styles and images, on 11 September 2026.

Page	HTML only	Full browser load	Requests	Multiplier
Wikipedia article	46 KB	424 KB	40	9.2×
Booking.com city page	106 KB	2,011 KB	108	19.0×
Hacker News front page	6 KB	not measured	1	n/a

On Booking the browser pulled 1,373 KB of JavaScript and 525 KB of stylesheets to render a page whose HTML was 106 KB, and that measurement ran with images blocked, so the real number is higher.

Now put traffic prices against it. Residential is $2.00 per GB at the entry package and reaches $1.00 per GB at volume, so:

What you fetch	Pages per GB	Cost per 1,000 pages
HTML only, 50 KB page	~21,000	$0.05 to $0.10
HTML only, 106 KB page	~9,900	$0.10 to $0.20
Rendered, 424 KB page	~2,470	$0.40 to $0.81
Rendered, 2 MB page	~510	$1.95 to $3.90

Read the last column next to what you can charge. At $0.10 per thousand requests, the market price of a plain managed request, a rendered 2 MB page costs you twenty times your selling price. The business is not broken by the target site or by your parser. It is broken by rendering pages you did not need to render.

Why is rendering the whole margin question?

Because the market prices it exactly the way our measurement does. ScrapingBee charges in credits, and the ladder is public: 1 credit for a plain request, 5 with JavaScript rendering, 10 with a premium proxy, 25 with both, 75 for stealth mode. That is a 75× spread on the same URL, published by a vendor who has no reason to exaggerate it.

So the first engineering question on any new target is not which library to use. It is: can I get this data without a browser?

Three things to try before you accept the 9× to 19× tax:

The site's own JSON. Most modern pages fetch their content from an internal endpoint, which you can call directly from Python. Open the network tab, find it, call it directly. You often get cleaner data than the HTML would give you, at a fraction of the bytes.
Server-rendered HTML. Plenty of sites still ship the content in the first response. Fetch it and parse it. Our Python crawling guide and Node guide both cover the parsing side.
Block assets when you do need a browser. Images, fonts, analytics and ad scripts are most of the weight and none of the data. Blocking them in Playwright or Puppeteer cuts a rendered page back toward its HTML size.

When is a managed API cheaper than running it yourself?

Honest answer: more often than a proxy vendor likes to admit.

Your situation	Cheaper option
HTML only, high volume, tolerant targets	Datacenter IPv4 at $1.15 to $1.35 per IP per month, unmetered traffic
HTML only, moderate volume, geo matters	Residential traffic, $1.00 to $2.00 per GB
Rendering required, low volume	A managed API: you are renting their browser farm, not just an IP
Rendering required, high volume	Your own headless fleet plus proxies, once the API bill passes the cost of the engineer maintaining it
Targets with hard anti-bot and no rendering escape	Managed API first. Measure, then decide whether to bring it in-house

The crossover is not a philosophy, it is a spreadsheet. Take your monthly page count, multiply by the measured page weight, price it at both ends, and add the engineer-hours the self-built path needs every month. The number tells you which column you belong in.

What does a 200 OK actually prove?

Nothing on its own, and this is where unit economics quietly die.

We requested an Amazon product page with a normal browser user agent. The response was HTTP 200, 3.8 KB, and its entire text was: "Amazon.com. Click the button below to continue shopping." No title, no product, no price. A success by every dashboard metric, and zero data.

Two consequences for a data business:

Measure success on content, not status. Assert that the field you came for exists: a price, a title, a non-empty list. Anything else is a failure, however cheerful the status code.
Your retry logic is a cost centre. A pipeline that retries on 5xx only will happily bill you for thousands of 200-coded bot walls. Retry on assertion failure, with a different address, and count those retries in your cost per thousand.

A related trap: pages that return real HTML but no content until scripts run. Our raw request to a Reddit listing came back 8 KB titled just "Reddit". That is a shell, not a page, and it is exactly the case where a browser is genuinely required.

How should you price the work?

Three shapes, and one rule.

Per record. Simple to sell, and you own all the volume risk. Only use it once you have measured the real cost per thousand on that specific target, including retries, and multiplied it by three.

Monthly retainer per feed. What serious buyers prefer, because their finance team wants a line item, not an invoice that jumps with your crawl volume. Price the retainer on the cost of keeping the feed alive, which is mostly maintenance when the target redesigns, not traffic.

Per field enrichment. The market anchor is $0.10 to $0.30 per contact for appended fields. Below that you are competing with providers who already have the data and sell it at scale.

The rule: never price a recurring feed on the cost of the first crawl. The first crawl is the cheap part. The expensive part is the Tuesday in March when the target changes its markup and your client's dashboard goes blank.

What actually kills these businesses?

Not blocking. Blocking is a solvable engineering problem with a known price.

Layout changes. Every target redesigns. A single client with five feeds will consume a day a month just staying alive.
One-off projects. A pipeline of one-time extractions is a consulting business with no compounding value. Convert everything you can into feeds.
Concentration. Two clients and 80% of revenue in one of them is not a business, it is a contract with a countdown.
Legal surprises. Personal data collected without a lawful basis, or resold onward, is the one failure mode that does not announce itself in a dashboard.
Pricing on last month's cost. Traffic prices fall and target defences rise. Re-measure your cost per thousand every quarter.

Where is the legal line?

Not legal advice, and the honest summary is shorter than most articles on the subject pretend.

Public pages are the safe ground. Collecting data that any visitor can see without logging in is the part with the strongest footing.
Logging in changes the analysis. Once you accept terms of service and authenticate, you are operating under an agreement you signed.
Personal data is regulated separately from access. Under GDPR it does not matter that a name and an email were public: you still need a lawful basis to process them, and the person retains rights over them. This is the piece that trips up list sellers, not the scraping itself.
Reselling other people's contact data drags the buyer in too. Broker registration rules exist in several jurisdictions, and downstream privacy claims land on whoever used the list.
Copyright survives extraction. Facts are not owned. A substantial copy of someone's text, images or database arrangement can be.

Practical version: sell derived numbers and public business data, treat personal data as a product decision rather than a technical one, and get actual counsel before building a business on a single legally aggressive source.

What would we do with $200 a month?

Because that is the honest starting budget, and it goes further than people expect.

$105 to $150 on traffic. At $1.00 to $2.00 per GB that is 75 to 150 GB, which is somewhere between 700,000 and 3 million HTML-only pages a month.
$0 on rendering until a target forces it, and then only for that target.
The rest on one paid data source you do not scrape. An official API for the part of your feed that has one is cheaper than defending a crawler against a site that offers the same data legitimately.

Two million pages a month is a real data product. Most people never get there, not because traffic is expensive, but because they render everything from day one and run out of money at 30,000 pages.

If your work is accounts rather than data, the arithmetic inverts and the per-account stack is priced in what a multi-accounting stack actually costs.

FAQ

How many pages can I scrape per gigabyte?

On HTML-only requests, roughly 10,000 to 21,000 pages per GB for pages in the 50 to 106 KB range we measured. With a full browser render, 500 to 2,500. The spread is the single biggest cost lever you control.

Do I need residential proxies for scraping?

Only where the target rejects hosting ranges. Datacenter IPv4 at $1.15 to $1.35 per address per month with unmetered traffic is dramatically cheaper for tolerant targets, and the proxy type should follow the target, not the other way round.

Is a scraping API a competitor to proxies or a customer of them?

Both. They buy proxies wholesale, add a browser farm and unblocking logic, and resell per request. That is a real service worth paying for when you need rendering at low volume, and an expensive middleman when you do not.

What is the fastest model to first revenue?

Done-for-you collection. One client with a concrete extraction pays this week, teaches you a target, and funds the feed you turn it into.

How do I know my scraper is actually working?

Assert on content, not on status codes. Our own test caught an Amazon page returning HTTP 200 with a 3.8 KB "continue shopping" wall and nothing else. Count those as failures and they stop quietly inflating your success rate.

Is scraping legal?

Collecting public data is generally on solid ground; authenticating, collecting personal data and reselling it are three separate questions with their own answers. Treat GDPR as a product constraint rather than a legal footnote, and get counsel before betting a business on one aggressive source.

How often should I re-measure costs?

Quarterly, and after any target redesign. Page weights drift upward, traffic prices drift downward, and a quote written a year ago is usually wrong in both directions.