How to clean scraped prices and dates with Playwright

To clean scraped prices and dates, read the browser’s resolved locale and timezone once per page, then parse every value against that single locale: numbers through a locale-aware library into Decimal, and relative or localized dates into UTC. Do not autodetect the format per value.

The hard part of scraping prices and dates is not extracting the text. It is turning that text into a typed number and a UTC timestamp when the text does not tell you, on its own, which convention it was written in.

1.234,56 is one thousand two hundred thirty four and change in one locale and is a syntax error in another. 03/04/2026 is in March or in April depending on where the page thinks you are. 2 days ago is not a date at all until you know the clock it was measured against. Every one of these depends on the locale the page rendered in, and that locale is decided by your exit, not by your parser. This page is how to parse them correctly, and why keeping the browser’s locale stable is what makes the parser reliable run after run.

Why locale is the hard part of price and date parsing

Three things vary with locale and all three break naive parsing:

  • The decimal and thousands separators swap. 1.234,56 and 1,234.56 are the same amount written two ways. A parser that assumes one reads the other off by a factor of a thousand, and it does it silently, because both strings are valid floats to a regex.
  • The currency symbol and its position move. Leading $, trailing ` kr, a non-breaking space, an ISO code like EUR` instead of a glyph. You have to strip the right thing before you parse the number, and what is right depends on the locale.
  • Dates arrive localized and relative. Month names in the page’s language, day and month in either order, and relative forms like 2 days ago, vor 2 Tagen, hace 2 dias. Resolving a relative form needs both the language and the clock it counts back from.

The trap is that all of this is stable only if the page keeps rendering in the same locale. Geotargeted pages render in the locale tied to your exit IP and timezone. If that pairing drifts between runs, the formats drift with it, and a parser that worked yesterday quietly misreads today.

Anchor the browser to its exit so the formats stay stable

Geotargeted content picks its locale from the exit the request appears to come from, cross-checked against the browser’s own timezone and language. Keep those consistent with the proxy exit and the page renders the same way every time, which is what makes a parser something you can write once. Let them disagree and you get two failures at once: the page may serve a different locale than you planned for, and the mismatch is itself a cross-check a detector uses to flag the session.

This wrapper derives the timezone and language from the egress IP by default, and a fixed seed makes the whole identity reproducible, so the locale you parse against does not move underneath you:

from invisible_playwright import InvisiblePlaywright

proxy = {
    "server": "socks5://gate.example.com:1080",
    "username": "user",
    "password": "pass",
}

with InvisiblePlaywright(seed=42, proxy=proxy) as browser:
    page = browser.new_page()
    page.goto("https://example.com/product")

    # Read the locale and timezone the browser actually ships. These are the
    # ones the page saw and rendered against, so they are the ones to parse with.
    locale = page.evaluate("() => Intl.DateTimeFormat().resolvedOptions().locale")
    tz = page.evaluate("() => Intl.DateTimeFormat().resolvedOptions().timeZone")
    print("parsing against", locale, tz)   # e.g. de-DE Europe/Berlin

    raw_price = page.inner_text(".price")   # "1.234,56 EUR"
    raw_posted = page.inner_text(".posted") # "vor 2 Tagen"

Because the browser returned by InvisiblePlaywright is a real Playwright Browser, page.evaluate, page.inner_text and every other method are the stock API. There is nothing wrapper-specific to learn on the extraction side. What the wrapper gives you is a locale and timezone that agree with the exit and stay put across runs, so the two values you read above are trustworthy inputs to the parser rather than a moving target.

Parse locale-formatted prices into typed numbers

Strip everything that is not part of the number, then parse the digits and separators with the page’s locale rather than guessing. Babel does the locale-aware step:

pip install babel

import re
from decimal import Decimal
from babel.numbers import parse_decimal

def clean_price(raw, locale):
    """'1.234,56 EUR' + 'de-DE' -> Decimal('1234.56')."""
    # keep only digits and the two separator characters and a sign
    number_part = re.sub(r"[^0-9.,-]", "", raw).strip(".,")
    # Babel wants an underscore locale ('de_DE'), the browser gives a hyphen
    value = parse_decimal(number_part, locale=locale.replace("-", "_"), strict=False)
    return Decimal(str(value))

price = clean_price(raw_price, locale)   # Decimal('1234.56')

parse_decimal reads 1.234,56 as 1234.56 under de_DE and 1,234.56 as the same under en_US, because it knows which character is the grouping separator in each. That is the entire point: the separators are ambiguous in isolation and unambiguous once you name the locale. Keep the currency symbol separately if you need it (a plain regex for the ISO code or glyph), and keep the amount as a Decimal, never a float, so cents do not drift.

Parse localized and relative dates into UTC timestamps

Relative and localized dates need the page’s language to read the words and the page’s timezone to anchor the arithmetic. dateparser takes both:

pip install dateparser

import dateparser

def clean_date(raw, locale, tz):
    """'vor 2 Tagen' + 'de-DE' + 'Europe/Berlin' -> aware UTC datetime."""
    language = locale.split("-")[0]          # 'de'
    return dateparser.parse(
        raw,
        languages=[language],                 # pin the language, do not sniff it
        settings={
            "TIMEZONE": tz,                   # the clock the page counted from
            "TO_TIMEZONE": "UTC",             # normalize everything to UTC
            "RETURN_AS_TIMEZONE_AWARE": True,
        },
    )

posted = clean_date(raw_posted, locale, tz)  # datetime in UTC, tz-aware

2 days ago only becomes a timestamp once you fix the timezone it was measured in; computed against the wrong zone it lands hours off, and near midnight it lands on the wrong day. Passing the same tz you read from the browser closes that gap. Store the result in UTC and keep the original string in a separate column, so a parse you later doubt can be re-derived instead of guessed.

Pin the parsing locale to the fingerprint, do not autodetect

The one rule that saves the most cleanup later: parse every value in a run against the one locale the browser reported, not against a locale you sniff per value.

Per-value autodetection looks convenient and fails on exactly the rows that matter. Feed an autodetector 1.234 and it has no way to know whether that is one thousand two hundred thirty four written in one convention or one-point-two-three-four in another; it will pick one, then pick the other three rows down, and your column ends up with a thousand-fold error scattered through it that no single value looks wrong enough to catch. The locale is not a property of the value. It is a property of the page, and you already read it once, up front, from the browser that rendered the page.

Concretely, in a batch of price strings mixing 1.234,56 and 1,234.56, per-value autodetection flips the decimal separator on every ambiguous short value it meets, while pinning to the page’s own locale reads all of them the same correct way. Read the locale once per page load, thread it through every clean_price and clean_date call, and the ambiguity is gone by construction.

This is also why a stable browser identity matters to a data pipeline and not just to staying unblocked. If the exit, timezone and language stay consistent, the locale you pinned at the top of the run is the locale every value on the page was written in.

A short demonstration

Put the three pieces together and one page yields typed rows:

from invisible_playwright import InvisiblePlaywright

with InvisiblePlaywright(seed=42, proxy=proxy) as browser:
    page = browser.new_page()
    page.goto("https://example.com/listing")

    locale = page.evaluate("() => Intl.DateTimeFormat().resolvedOptions().locale")
    tz = page.evaluate("() => Intl.DateTimeFormat().resolvedOptions().timeZone")

    rows = []
    for card in page.query_selector_all(".card"):
        rows.append({
            "price": clean_price(card.query_selector(".price").inner_text(), locale),
            "posted_utc": clean_date(card.query_selector(".posted").inner_text(), locale, tz),
            "price_raw": card.query_selector(".price").inner_text(),
        })

    for r in rows:
        print(r["price"], r["posted_utc"].isoformat())

Re-run it with the same seed and proxy and the page renders in the same locale, so the same parser reads the same rows the same way. That reproducibility is the difference between a cleaning step you trust and one you re-audit every morning.

Conclusion

Cleaning scraped prices and dates is a locale problem wearing a parsing costume. Read the browser’s own locale and timezone once per page, parse numbers with a locale-aware library instead of a bare float, resolve relative and localized dates against the timezone the page used, keep amounts as Decimal and timestamps as UTC, and never let a per-value autodetector guess a separator it cannot know. The upstream requirement for all of it is a browser whose locale and timezone stay consistent with the exit, because that is what keeps the formats stable enough to parse the same way twice.

Short answers to the questions that lead here

Why is my scraped price a thousand times too big or too small? Your parser assumed the wrong decimal separator. 1.234,56 and 1,234.56 are the same amount in different locales; parse with the page’s locale instead of a bare float().

How do I parse “2 days ago” into a real date? With a library that handles relative dates, and pass it the page’s timezone as the base clock. Without the timezone it lands hours off and, near midnight, on the wrong day.

Should I autodetect the locale for each value? No. Pin one locale per page, read from the browser, and use it for every value. Ambiguous short numbers are exactly the ones autodetection gets wrong.

Why do the formats change between runs? Because geotargeted pages render in the locale tied to your exit, and if the exit, timezone or language drift, the formats drift with them. Keeping them consistent with the exit keeps the formats stable.

Do I need the currency symbol to parse the number? No, strip it first. Keep it in a separate field if you want it; it does not belong in the numeric column.

Float or Decimal for money? Decimal. Floats lose cents at scale, and money that is off by a rounding error is off.

Sources

  • The wrapper’s real API for launching the browser and reading the page, from the Quickstart and Configuration pages in this set.
  • Intl.DateTimeFormat().resolvedOptions() for the browser’s resolved locale and timezone, read live from the page rather than assumed.
  • The locale-aware behaviour of the Babel and dateparser libraries, used as documented upstream.

See also: how to scrape geotargeted content for making the exit, locale and timezone agree in the first place, when the timezone does not match the proxy for what a mismatch costs beyond broken formats, and how to export scraped data to CSV for keeping the Decimal and UTC values intact on the way out.

Written while maintaining invisible_playwright, a Firefox patched at the C++ level driven by stock Playwright. The locale you parse against is a property of the page, and the page’s locale is a property of your exit.

Back to top

MIT licensed. Every page here is written against the current source of the thing it describes, and several record something we got wrong first.

This site uses Just the Docs, a documentation theme for Jekyll.