convexworks/cookie-banners

Cookie Banner Text and Controls

The wording, controls and markup of 1,010 cookie banners, taken from the saved homepages of European B2B software companies in June 2026. Each row is one banner: its title, its text, every button, link and toggle on its first screen, and the banner's HTML, in the original wording.

Use it to study how consent is worded, or to prototype tools that read banner text and button labels. It does not measure legal compliance.

Prepared by Convex Works from the European Software Data Study.

Quick start

import pandas as pd
from datasets import load_dataset

repo = "convexworks/cookie-banners"
banners = load_dataset(repo, "banners", split="train").to_pandas()
judgments = load_dataset(repo, "model_judgments", split="train").to_pandas()
sites = load_dataset(repo, "sites", split="train").to_pandas()
df = banners.merge(judgments, on="domain", validate="one_to_one").merge(
    sites[["domain", "vertical", "business_model"]], on="domain", validate="one_to_one"
)

# One row per control, keeping the domain it belongs to.
controls = pd.DataFrame(
    [{"domain": row.domain, "consent_tool": row.consent_tool, **c} for row in df.itertuples() for c in row.controls]
)
buttons = controls[controls.kind == "button"].dropna(subset=["label"])

# Where the model found no direct reject button, what do the buttons say?
no_reject = df.loc[df.reject_button == "not_detected", "domain"]
print(buttons[buttons.domain.isin(no_reject)].label.str.casefold().value_counts().head(8))

label
accept                55
accept all            36
cookie settings       24
settings              16
accept all cookies    15
ok                    12
learn more            11
customize             10

Many sites use a consent tool's default wording: the 1,010 banners have 701 distinct wordings. Use text_id to count each wording once.

Tables

Config	Rows	One row per
banners (default)	1,010	banner
model_judgments	1,010	banner: the model's reading of its accept and reject buttons
sites	2,357	homepage in this version: its segment, and whether its banner is in banners

Every table joins on domain, the company's domain as listed in the source directory. Each table has one row per domain.

Versions and segments

Version 1.0 covers companies classified as B2B software (vertical = software, business_model = b2b or b2b2c). Later versions add other segments as new rows in the same tables, so code written against 1.0 keeps working. Filter on the segment columns in sites to keep a fixed population, and pin a version with load_dataset(repo, "banners", revision="v1.0.0").

Fields

banners

Field	Type	Description
domain	string	Join key: the company's domain as listed
page_url	string	URL of the saved homepage after redirects
captured_at	timestamp (UTC)	When the page was saved
page_language	string, nullable	The page's lang attribute as written, such as en-GB. Null when missing. The banner itself may use another language.
consent_tool	string, nullable	Consent tool recognised from the banner's markup; its first screen was located with that tool's own markup. Null means none of the eight tools listed under How it was made, and the whole banner element was used; the banner may still come from another tool.
title	string, nullable	Banner heading, when the consent tool marks one. Other banners keep any heading as the first line of text.
text	string	Banner wording, one line per paragraph. Standalone buttons and links are removed; links inside a sentence stay.
controls	list of struct	Every control on the first screen, in page order
html	string	The banner's HTML in the same scope as text and controls: later layers, elements marked hidden and hidden form inputs are removed, as are scripts, styles and SVG; embedded data: images are shortened. Otherwise as saved, including classes, roles and inline styles.
text_id	string	Hash of title and text. Identical wording gives the same id.

Each control:

Field	Description
kind	button, link or toggle, following the HTML. An "Accept" built as a plain link is a link.
label	Visible text. Null when the control shows no text, such as a close icon.
aria_label	The aria-label attribute, if any
element	HTML tag, such as button, a, input or div
role	The role attribute, if any
checked_in_html	Toggles only. The saved HTML marks the toggle as on (checked, or aria-checked/aria-pressed="true"). Scripts can change a toggle without updating this markup, so it does not show what the visitor saw.
disabled_in_html	The saved HTML marks the control as disabled (disabled or aria-disabled="true")
href	Links only. Absolute target URL; null for in-page or script links.

model_judgments

Field	Description
domain	Join key
accept_button	detected or not_detected: whether the model found a direct accept button among the captured labels
reject_button	The same for a direct reject button, such as Reject, Deny, Decline or necessary-only. Settings and Manage buttons do not count.
rechecked	true when the first answer failed a consistency check against the captured labels and the model answered again with high reasoning; the second answer is the one shown

Model mistral-medium-2604, prompt version privacy_banner_saved_html_core_v7; every answer had confidence "high". The model read the original extracted fragments, not the cleaned rows in banners. not_detected means not found among the captured labels, not that the site has no such button. No person reviewed these judgments.

sites

Field	Description
domain	Join key
page_url	URL of the saved homepage after redirects
captured_at	When the homepage was saved (UTC)
consent_tools_in_page	Consent tools whose code appears anywhere in the saved page, found by fixed text markers, whether or not a banner was captured. 335 sites with no_banner_found load one.
vertical	Segment: what the company sells, such as software
business_model	Segment: who pays, such as b2b or b2b2c
offering_type	Segment: the form of the offer, such as software_product or infrastructure_or_connectivity
status	released (1,010), excluded (169), no_banner_found (1,090) or unclear (88)
reason	Exclusion reason when status is excluded, the first check the banner failed; otherwise null

The segment columns come from gpt-5.4-nano, one pass over each company's homepage, about, pricing and demo text. They were not reviewed by a person. no_banner_found and unclear are the banner model's reading of the saved HTML. They do not show that the site had no banner.

How it was made

1. Pages. Homepages of 2,357 companies listed in the EU-Startups directory, classified as B2B software and with estimated search traffic, saved as rendered HTML by a headless browser on 21–22 June 2026. No page was fetched again for this release.
2. Candidates. A script collected page fragments with cookie or consent wording, together with their controls.
3. Banner or not. mistral-medium-2604 judged whether each homepage's fragments contained a sitewide cookie banner: 1,179 yes, 1,090 no, 88 unclear. Answers that failed a consistency check against the captured labels were asked again with high reasoning.
4. First screen. For eight consent tools (Cookiebot, CookieYes, HubSpot, Complianz, Finsweet Cookie Consent, iubenda, OneTrust and CookieScript), the first screen is located with the tool's own markup. Later layers such as settings panels, vendor lists and detail tabs are left out. For Complianz, the hidden parts follow its default stylesheet. Other banners use the whole banner element.
5. Cleaning. Elements the HTML marks as hidden (the hidden attribute, inline display: none or visibility: hidden, or the tool's hide class) and screen-reader-only text are left out. Whitespace is normalised; the wording is otherwise unchanged. Emails and phone numbers would be replaced with [email] and [phone]; none remained in the released rows.
6. Selection. 1,010 of the 1,179 banners pass every check. The rules were set by inspecting examples. No person checked the released rows.

Exclusion reason	Sites	Meaning
banner_hidden_in_saved_html	32	The banner was marked hidden when the page was saved
repeated_controls	28	The same control appears twice, from a second layer or a duplicate layout
contains_tabs_or_form_fields	23	Tabs, dropdowns or form fields inside the banner
text_too_long	16	Over 1,500 characters, with page or settings text mixed in
contains_page_content	15	The banner element holds navigation, main page content or over 50,000 characters of markup
too_many_controls	15	More than 12 controls
unlabelled_toggle	11	A toggle with no label
no_consent_wording	8	No cookie or consent wording
control_label_too_long	8	A control label over 60 characters
multiple_banner_candidates	6	More than one possible banner element
unsupported_tool_layout	5	A Complianz layout the rules do not cover
extraction_not_reproducible	2	Re-extraction did not match the original evidence

Limitations

• Saved HTML is not a screen recording. It does not show what was visible, how prominent each button was, or what clicking did. Stylesheets loaded from separate files were not available.
• One capture per site, from one location, at one time.
• Version 1.0 covers B2B software companies listed in EU-Startups with estimated search traffic, not a sample of all websites.
• kind follows the HTML: accept or reject options built as links appear as link.
• Banner detection, the accept/reject judgments and the segment columns come from model passes and were not reviewed by a person. Thirteen homepages contain the banner markup of one of the eight known consent tools but were not marked as having a banner (7 no_banner_found, 6 unclear).
• The dataset is not evidence of legal compliance or non-compliance.

Licence and rights

Our selection, structure and labels are released under CC BY 4.0. The banner wording is quoted from the websites listed; rights in it remain with the site owners and consent-tool vendors, and CC BY 4.0 does not cover it.

Corrections and removal

Site owners can ask for a row to be corrected or removed by writing to hello@convex.works or opening a discussion in this dataset's Community tab. Changes are applied in the next version.

Citation

@misc{convexworks_cookie_banners_2026,
  title  = {Cookie Banner Text and Controls},
  author = {{Convex Works}},
  year   = {2026},
  url    = {https://huggingface.co/datasets/convexworks/cookie-banners},
  note   = {Version 1.0.0}
}

Version

1.0.0 (October 2026): first release, B2B software companies.