First-party cookies are set by the website you are visiting, and third-party cookies are set by a different domain embedded in that page. The first kind keeps your session working: your login, your cart, your preferences. The second kind follows you across websites so ad networks can recognize the same person and measure what they do. That difference, same-site or cross-site, is the root of the privacy changes, browser restrictions and tracking problems that followed.

Most explanations of this topic were written years ago and still talk about an impending “phase-out” that never landed the way people expected. The current picture is more specific. Third-party cookies still work in regular Chrome. Safari and Firefox have neutralized them for cross-site tracking for years. And Google walked back its own plan to remove them.

I set up and retire tracking for clients and for our own sites, and the cookie details below are the ones that decide whether a setup keeps its data. This guide covers what the two kinds of cookies are, how each browser treats them today, what Google decided, and what that means for the data your ad platforms receive.

A cookie is a small piece of data stored in the browser. What makes it first-party or third-party is not which company set it. It is the site relationship at the moment the cookie is set or read.

First-party cookies are set in a same-site context: the domain that sets the cookie belongs to the same site as the page you are on. The cookie that keeps you logged in, set by the site you signed into, is first-party. These handle logins, language choices, saved preferences and session continuity. Because they stay within the site that set them, browsers generally treat them as lower-risk.

Third-party cookies are set in a cross-site context: the domain that sets the cookie belongs to a different site than the page in your address bar. They usually arrive through embedded content like an ad, an analytics tag, a social widget or a tracking pixel. Because the same external domain can load on thousands of sites, its cookie can read the same identifier on each one and stitch your movement together. The cookie an ad network sets while you read a news article, so it can show you a related ad on an unrelated site later, is the classic example.

The mechanism is identical. Both are the same kind of file, stored and sent by the browser. The classification follows a non-obvious rule: the same vendor’s cookie, served from your own subdomain, is treated as first-party. Served from the vendor’s own domain on someone else’s page, it is third-party. The browser looks at the site relationship, not at the vendor behind it.

Flow diagram: the same cookie being set as first-party is kept by the browser, while as third-party it is blocked.
Nothing about the visitor changed. What changed is which domain set the cookie.

First-party vs third-party cookies side by side

First-party cookies Third-party cookies
Who sets it The site in the address bar A different, embedded domain
Typical reach A single website Across many websites
Common use Login, cart, preferences Retargeting, ad measurement, profiling
Safari Works; cookies set by JavaScript are deleted after 7 days without interaction, or capped at 24 hours when ITP detects qualifying link decoration Blocked by default; CHIPS cookies stay partitioned by top-level site, and the Storage Access API can grant an embed access to its unpartitioned cookies
Firefox Works Partitioned per site by default (desktop, since 2022); known trackers blocked
Chrome Works Allowed by default; blocked in Incognito; users can restrict
Privacy law exposure Lower, but not exempt Higher
Role in conversion tracking Carries session and click identifiers Unreliable for cross-site identity

The first-party cookie is what you keep. The third-party cookie is what the browsers have been taking away, at different speeds and by different methods.

How each browser treats cookies today

The browsers do not behave the same way, and mixing them up is the most common mistake in this conversation.

Safari blocks third-party cookies by default and limits first-party ones too. WebKit’s tracking prevention page (checked 2026-10-17) says ITP “by default blocks all third-party cookies” and that access can only be granted through the Storage Access API. A Storage Access API grant gives an embedded site its unpartitioned cookies, the same ones it sees elsewhere, on the sites where the visitor allowed it. Safari 26.2 brought back partitioned cookies (CHIPS), which are isolated per top-level site and so cannot carry an identity between sites.

The first-party rules matter more for tracking. ITP deletes all cookies created in JavaScript, and other script-writable storage, after 7 days without user interaction with the site, whatever the cookie is used for. When ITP detects link decoration, it caps the JavaScript cookies set on that landing page at 24 hours. WebKit’s ITP 2.2 post gives the conditions: a domain classified with cross-site tracking capabilities navigated the visitor there, and the final URL has a query string or fragment. _ga, _fbp and _fbc set by a browser tag fall under those rules.

Firefox partitions third-party cookies rather than refusing them. Mozilla made Total Cookie Protection the default for all Firefox desktop users in June 2022. Each website gets its own cookie jar, so a cookie set on one site cannot be read on another. Enhanced Tracking Protection additionally blocks known trackers from a list. The effect is close to a block for cross-site tracking, but the mechanism is isolation plus tracker blocking.

Chrome still allows third-party cookies by default. In regular browsing they go through. Incognito blocks them by default, and users can restrict them in settings. That is why Chrome is the browser where third-party cookies still work for most visitors.

A visitor on Safari or Firefox is already largely opaque to cross-site tracking through third-party cookies. A visitor on regular Chrome is not.

What actually happened to third-party cookies

The “death of the third-party cookie” is a real story, but most tellings stop at the original plan and skip the reversals. The timeline has two halves.

The first half is the browser restrictions that already happened. Safari 13.1 shipped full third-party cookie blocking in March 2020, and Firefox made Total Cookie Protection the desktop default in 2022. For users on those browsers, third-party cookies stopped working for cross-site tracking years ago.

The second half is the Chrome phase-out that never landed. Chrome planned to deprecate third-party cookies entirely, and that plan produced years of “cookies are dying” coverage. In July 2024, Google said it would offer users a choice instead of removing them. In April 2025, it said it would keep its current approach and dropped the standalone choice prompt it had proposed. On 17 October 2025, Google announced it would retire most of the Privacy Sandbox technologies meant to replace third-party cookies, including Topics, Protected Audience and Attribution Reporting, through Chrome’s phased deprecation process. It kept CHIPS, FedCM and Private State Tokens.

So in 2026, third-party cookies still function in regular Chrome, while Safari and Firefox neutralize them for cross-site tracking. The loss of cross-site tracking is real and has been for a long time on the browsers a large share of visitors use, especially on mobile and among audiences that skew toward Apple devices.

The takeaway for tracking does not depend on which headline wins. Relying on third-party cookies means accepting that a large portion of your visitors run browsers that defeat them. Building on first-party data means your measurement leans on identifiers those browsers still allow.

Blocking third-party cookies weakens cross-site identity, not every path

From a visitor’s point of view, blocking third-party cookies reduces cross-site tracking. Ad networks have a harder time following you between unrelated sites, and the retargeting that depends on that weakens. Most sites keep working, because logins and carts run on first-party cookies. Some embedded tools, like social login buttons or certain external video players, may not work until you allow third-party cookies for that site.

From a marketer’s point of view, the effect is narrower than “tracking breaks.” Google stores the click identifier in a first-party cookie through its tag and Conversion Linker, Meta offers first-party cookie options and the Conversions API, and platforms apply conversion modeling to fill gaps. What third-party cookie loss mostly removes is reliable cross-site recognition and view-through measurement, not all attribution.

What this means for conversion tracking

The platforms need to connect two moments: the ad click and the conversion that followed. Third-party cookies were one way to bridge them for cross-site journeys, and for many conversions they have not been the main one for a while. The impact differs by use case.

Remarketing audiences shift toward first-party events. As the cross-site cookie weakens, remarketing depends more on site events you collect and send to the platform, on customer lists and on platform-side identity resolution. The audience gets smaller and more dependent on data you own.

Click attribution leans on first-party click IDs. Google stores the click identifier in a first-party cookie on your domain, so the link between a click and a later conversion can survive without third-party cookies. In Safari, that cookie is still subject to the 7-day and 24-hour rules above when a script sets it. The gap shows up when the identifier is missing: after a consent decline, a cross-device journey, a week without a return visit in Safari, or a setup that never captured it.

Match quality and view-through measurement weaken. With fewer cross-site identifiers, platforms recognize fewer people across the journey, and view-through attribution gets noisier. Conversion modeling partially compensates, but it is an estimate layered on thinner signal.

The durable response is to collect the data you can rely on in a first-party context, send it through your own infrastructure, and accept that some cross-site signal is gone for good.

Why first-party collection lasts longer, and where it stops

Three ideas get bundled together here. First-party data is anything you collect directly from your own audience, whether it lands in a cookie, a CRM or a database. A first-party cookie is one storage mechanism for that relationship. A same-site server endpoint is a way to move the data so the browser treats it as part of your site. When you run a server-side tagging container on a subdomain like data.yourdomain.com, the browser sees same-site traffic, and the endpoint is not on the static blocklists that target known tracking domains.

That is an improvement, not guaranteed evasion. Advanced filters and extensions can target custom endpoints. Safari also checks how your subdomain resolves. WebKit says ITP “detects third-party CNAME cloaking and third-party IP address cloaking requests and caps the expiry of any cookies set in the HTTP response to 7 days”. A subdomain that is a CNAME to a hosting provider, or an A record pointing at a third-party IP, can land in that cap even though the name looks like yours. Same-site naming is not the same as a full cookie lifetime, so check how your endpoint actually resolves.

It helps to be precise about what moves where. The cookie stays in the browser, the browser stores it and sends it back, and a Set-Cookie header only instructs the browser. The server can hold session state linked to an identifier, and it sends events to each platform through its API. Nothing relocates the cookie itself to the server.

A server-managed FPID configuration can also use a different client ID from the one your browser tags use. Setting a cookie over HTTP doesn’t do that by itself; the value the server chooses does. On a travel client’s server container I took apart in October 2026, the GA4 client set an HttpOnly FPID cookie with a 400-day max age, and on 2026-10-09 the client ID it forwarded to GA4 differed from the one the browser’s Google tag held in its own JavaScript cookie. A configured max age is a request to the browser, not observed persistence, and I have not yet published a Safari measurement of how long that cookie actually survived. The server-side GTM guide covers what that ID split means when you leave a container.

ATA’s server GTM containers run on this kind of first-party endpoint you control. A managed host like Stape, where I run client containers, handles the hosting, the custom domain and the scaling.

First-party, server-side collection improves how completely your own events reach each platform. It does not rebuild cross-site identity, restore what a consent decline removes, or replace audience reach that depended on cross-site tracking. What you are allowed to collect reaches the platforms more often, and it still has to be monitored and reconciled against your orders.

What first-party collection does not fix

Consent still applies. A first-party, server-side setup is not a way to route data around a declined consent banner. GDPR, ePrivacy and CCPA expect you to honor the visitor’s choice, and Google Consent Mode is built around that signal. Being first-party is not a legal exemption. Strictly necessary cookies get different treatment, but first-party analytics and advertising cookies can still require consent in the EU, and CCPA governs collection, sharing and sale with notice and opt-out rights.

Cross-site identity does not come back for free. First-party data is tied to your domain, so it does not recognize the same person on another site the way an ad-network cookie did. Cross-site reach comes back through consented customer matching, hashed identifiers and platform-side identity graphs.

Safari’s limits apply to first-party cookies as well. JavaScript-set cookies expire after 7 days without interaction, and server-set cookies get the same 7-day cap when the endpoint looks like CNAME or IP cloaking. Measure what your visitors’ browsers actually keep instead of assuming the configured lifetime. To see the cookie lifetime each tracking setup gets in Safari and test yours, read how long cookies last in Safari.

Garbage in is still garbage out. If your browser data layer double-fires events, moving the endpoint to a server forwards two broken events instead of one. That is why ATA Edge validates purchases against a confirmed payment webhook instead of trusting whatever the page fires.

Where to go from here

For the architecture behind first-party collection, start with the server-side tracking guide. The Facebook Conversions API guide explains the server event, customer matching and the limits cookies do not solve. When you are ready to implement, the Meta Conversions API with GTM guide covers the container, deduplication and webhook validation.

The ATA Tracking Toolkit membership includes ATA Edge, for first-party server-side delivery to Meta with webhook validation, and a library of server GTM containers as a bonus.