Privacy

1. What is collected

When you sign in with Google

Signing in is also signing up: a new account starts with 10 free credits. To sign you in, Google is asked for three scopes — openid, your email address, and your basic profile — and nothing else. Passcite has no access to your Gmail, Drive, contacts or calendar. From that, these are stored:

• Your name, email address, and the URL of your Google profile picture.
• A Google access token and ID token, kept so the sign-in works. No refresh token is requested for signing in, so Passcite cannot act on your Google account when you are not using it.
• A session record: a random token, the IP address and browser user-agent the session was created from, and an expiry 7 days out, moved forward on each day you use it.

When you sign in with an emailed code

You type an email address and Passcite sends a six-digit sign-in code to it. The code works once, expires after 15 minutes, and stops working after three wrong tries. From that, these are stored:

• Your email address, as the account. No name and no picture — the code proves the address is yours and nothing more.
• A hashed copy of the code, deleted when it is used or expires. The code itself exists only in the email.
• The same session record as above.

When you set up a brand

A brand is one website you track — your own, or a client's. You give its address, and PassciteBot reads its public pages to suggest a profile: what the business is, what it sells, where its buyers are, the language they ask in, and businesses it competes with. It reads only what the site's robots.txt allows (see /bot). What is kept is the profile as you confirm or correct it, the competitors you keep, whether you work on your own brand or for clients, and the prompts you track: the questions a buyer would ask, suggested from the profile or typed by you.

When a report runs

Each prompt is asked of the AI engines, and each answer is stored with the brand: its text, the model that gave it, whether and where the brand was named, who was named instead, the pages it cited, and the search words the engine used. After a report, the site is read again to suggest tasks — what to change so the brand is named — and the answers to prompts that name the brand are read for Brand Audit.

An answer to a prompt that names no business can be shown again, for up to 7 days, to another account that tracks the same prompt on the same engine — it is the engine's public answer to a general question. An answer to a prompt that names a brand is never reused, and nothing about who tracks a prompt goes with the answer.

When you write and publish

An article is written from the brand's site, its profile and the facts you add to its knowledge base — for a comparison, also from the other business's homepage. Stored with it: each version, your edits and the instructions you gave for a rewrite, what the writer left out for lack of a source, when it was approved and published and the address it went live at, whether it was seen on that page, the answers to its prompts asked again a week later (Proof), and whether Google had the page when they were asked — from Search Console when it is connected, otherwise from a Google search for the page's address through SerpApi. Autopilot keeps its schedule, and — if you turn it on — the date you allowed it to publish new pages without asking, which is recorded on every page it publishes.

When you look at sources, research or check a site

The pages an engine cites in its answers are read by PassciteBot, as its robots.txt allows, for their title, their author and which of the brand's competitors they name; for a YouTube video, YouTube's public oEmbed endpoint is asked for its channel. What was read is kept with the brand and read again after 30 days. The pages you follow under Tracked pages are kept as addresses.

A research keeps the keyword or the page address you gave it and the questions it found; a page is read by PassciteBot first. A site health check reads the page you name — on the brand's site, or a competitor's homepage — with its site's robots.txt, llms.txt and sitemap, and Google PageSpeed Insights opens the page to score it; the findings are kept with the brand, and what needs changing on the brand's own site is added to its tasks.

When you share or deliver a report

A report, a client report or an article can be given a link that opens without signing in. The link carries a random key; anyone who has it can open what it points to, until you close it or the end date you gave it passes. For each link, the number of times it was opened and when it was first and last opened are kept — not who opened it, and not their address. Link previews made by chat and mail apps, and your own visits while signed in, are not counted. A shared page loads nothing from anywhere else.

A client report keeps every edition as it was when it was prepared: its figures, its lists and the note you wrote. A schedule keeps when it asks and which prompts. On the Agency plans you can add up to 10 addresses a client report is emailed to; those addresses are stored with the report, and each edition records who it was sent to. White label stores the agency name, logo, colour, tagline and contact line you enter; the logo is kept in our database and shown from there.

A site owner sent an article's link can tell us the address where it went live. That address is kept on the article; nothing about the person who gave it is.

When you prospect

A market is a kind of business in a place that you enter. Scanning it asks the engines questions that name no business, and its answers are stored like any report's. The businesses of that kind the map lists there — name, website, rating, review count, address and category — are fetched through SerpApi, which reads Google Maps, and kept in our cache for up to 30 days; a business you add to your list keeps only its name and website.

For a business on your list, PassciteBot reads its homepage — and its contact page when the homepage shows no address — for the email address, phone number and contact page the business publishes on its own site, what the site says it is, the firm its footer credits and what it is built with. It reads only what the site's robots.txt allows; nothing is guessed, decoded or looked up anywhere else. A picture of the homepage and its speed on a phone come from Google PageSpeed Insights, which opens the page itself. An address you type for a business is kept as yours.

A pitch reads the business's site the way a brand's is read, asks its prompts and stores the answers. Its email is drafted from the pitch's numbers and kept with the prospect; you send it from your own mail app or Gmail, or by copying it, and Passcite never emails a prospect. What is recorded is the date you sent it, the date of the one follow-up, and whether its link was opened. The agency name, contact line and your name for the sign-off are kept from Pitch settings. Marked won, the business becomes one of your brands, and its pitch its first report.

When you do outreach

A page AI answers cite, or one you add, is kept with the brand the first time you act on it. To find whom to write to, PassciteBot reads the page's byline, and its site's homepage — and contact page when the homepage shows no address — for the email address or contact page the site publishes, as its robots.txt allows; nothing is guessed, decoded or looked up anywhere else. An address you type is kept as yours.

An email is drafted from the brand's own facts and what the page says, read at that moment and not kept, and the email is kept with the page; you send it from your own mail app, Gmail or Outlook, or by copying it, and Passcite never sends it. What is recorded is the date you sent it, the date of the one follow-up, and that they replied, with a note if you write one. A check reads the page again for the brand's name and its links, and keeps what it found.

A reply to a community thread is drafted from the brand's facts and the thread's text — the text you paste, which is kept, or the page as its robots.txt allows — and Passcite never posts it. The link to your reply is kept if you give it.

When you connect WordPress, Webflow or Google

Each connection belongs to one brand. Disconnecting deletes it.

• WordPress: the site address, a WordPress username and an application password created for Passcite, the password stored encrypted. Used to check what it is allowed to do, to publish the new pages you approve, and to put one back to a draft when you take it down.
• Webflow: an API token created in that site's settings, stored encrypted, with which site and CMS collection new pages go into. Used the same way.
• Google Search Console: a separate Google consent asking for one scope, webmasters.readonly, which can read Search Console and change nothing. The refresh token it returns is stored encrypted with the property it reads. The figures are read when you open the brand's Proof, Website analytics or Your Google demand page, and shown, not stored — with four exceptions. Indexing keeps, for each page of the brand's sitemap it follows (up to 500) and each article published from Passcite, whether Google has it, Google's reason when it does not, when Google last read it and the link to it in Search Console, so each page is asked about at most once a day; and when you asked Google to index it from here. Disconnecting deletes them, and so does removing the brand. When an article's Proof asks its prompts again, the article keeps whether Google had its page that day. A client report's edition keeps the Search Console figures as they were when it was prepared — the last 28 days' clicks and impressions, the top five searches, and whether Google has indexed the pages that went live — so the report you send does not change afterwards, and it shows them to the people you send it to. And a search you track from Your Google demand becomes one of the brand's prompts.
• Google Analytics: another separate consent, for one scope, analytics.readonly, which can read Google Analytics and change nothing. The refresh token is stored encrypted, with the property that is the brand's site — matched by its web stream, or picked by you from the ones the account can read. Visits by day, landing page, channel and source are read when you open Website analytics or AI traffic, and shown, not stored.

Passcite's use of information received from Google APIs adheres to the Google API Services User Data Policy, including its Limited Use requirements. What Search Console and Google Analytics return is used only for the features described above: the pages that show it to you for the brand you connected and, for Search Console, Indexing and the tasks it files, an article's Proof, the client reports you send, and the searches you choose to track or research. It is not used for advertising, not sold, not passed to anyone except as described here — the people you send a client report to; for a search you track, the AI engines that answer it as a prompt; and for a search you research, Anthropic's model, which writes questions from it — not used to train AI models, and not read by a person unless you ask for help with it or the law requires it. How it is protected is in section 6.

When you pay

Checkout happens on Dodo Payments' own page, and the card goes to Dodo, never to Passcite. Passcite sends Dodo your email address, your name if the account has one, the account's ID and what you are buying; Dodo sends back its customer, subscription and payment IDs and the subscription's status and dates, which are stored against your account. Your invoices are read from Dodo when you open Subscription, not copied. Every credit given, bought, spent, refunded or lapsed is a row in your account's ledger, with the brand it was spent on — that is what Usage lists.

When a limit is counted

A few doors are counted in Upstash Redis, and the counters expire with their window: the free tools, 60 requests an hour per IP address; sign-in codes, 10 an hour per IP address and 3 per inbox in five minutes, and tries at a code, 10 an hour and 30 a day per inbox and 30 an hour per IP address — keyed on the address with any +tag (and, for Gmail, any dots) taken out; connecting a WordPress or Webflow site, 10 attempts an hour per account; reading a brand's site to fill in its profile, 30 a day per account; writing a brand's suggested prompts in its buyers' language, 60 a day per account; asking to be listed in the agency directory, 5 an hour per IP address; and on shared links, 30 misses (a link that does not exist) and 10 live-page answers an hour per IP address.

When an agency is in the directory

The agency directory lists agencies that say on their own site that they do AI search work. Each entry is what the agency publishes there — its name, website, the cities of its offices, what it says it does and for which industries, the year it says it started and its contact page — with the page it was read from and the day it was checked, and a small picture of its homepage as Google PageSpeed Insights loaded it. To ask whether an entry is right, we also keep the email address or contact form the agency's own site publishes and write to it once, with one follow-up, keeping what we sent and any reply. Nothing else is kept about anyone listed. To correct or remove an entry, write to hello@passcite.com.

An agency that asks to be listed sends its name, site, city and an email address to reply to, and if it likes a note. That is kept in our database, read only to answer the request, and deleted once it has been answered. The badge an agency can put on its site is an image from passcite.com; we do not count or record who loads it.

When you connect an AI assistant

You can let Claude, ChatGPT, Cursor or another assistant into your account from Settings → MCP, after signing in and allowing it. Stored for that: the assistant as it describes itself (its name, website and the address it returns you to), that you allowed it and when, its refresh tokens as a hash, the keys its access tokens are signed with (the private one encrypted), and when it last called. What it reads, it reads when you ask it to; see section 5 for where that goes.

When you open any page

Page visits are counted with Vercel Web Analytics. It sets no cookie and reads nothing already stored on your device. A visitor is identified by a hash Vercel computes from the incoming request, and that identifier is discarded after 24 hours, so there is nothing that could link today's visit to next week's.

Per visit it records the time, the path, the site that linked here, any query parameters, an approximate location from the IP address, the device type, and the browser and operating system versions. In production the script and the address it reports to are both on passcite.com — nothing new is fetched from another host, and the figures reach Vercel because Vercel already runs the servers that answer every request here.

The signed-in product is excluded. Pages about a brand (/b/…), your brands, the businesses you prospect, settings and the setup conversation, the admin console, shared links (/s/…, where the link itself is the key), and the pages where you let an assistant in are discarded in your browser and never sent, and neither is the address a sign-in page returns you to.

2. What is not collected

• No ad tracking, and no tracking across sites. There is no tag manager, no advertising pixel and no third-party cookie anywhere on this site. Page visits are counted as described above, by the company that already serves this site; they cannot identify you, are not sold or shared, and cannot follow you anywhere else.
• Three things load from elsewhere. The signed-in pages show the icon of each site in your account — your brands, their competitors, the sites the answers cite, and the businesses you prospect — and the agency directory shows each listed agency's, all from Google's public favicon endpoint, so Google sees the IP address of whoever opens one of those pages and which domain the icon is for. Signed in with Google, your profile photo is loaded from Google. The Open Graph preview tool shows the image the checked page names, loaded from wherever that page says. None of our cookies travel with any of them.
• Two cookies of our own, both for signing in. One holds the session token, is marked Secure, SameSite=Lax and HttpOnly — scripts on the page cannot read it — is not shared with subdomains, and lasts 7 days from the last time you used it. The other remembers which way you last signed in — Google or an emailed code — for 30 days, so the sign-in page can mark it; it holds that and nothing else. Signing in with Google, and connecting Search Console or Google Analytics, also set a short-lived cookie, gone within ten minutes, that ties Google's answer to the request that asked for it. This browser's own storage keeps one preference: whether the side column is collapsed.
• No account passwords. Sign-in is Google or a code emailed to you, so there is none to store; the code is kept only as a hash until it is used or expires. (A WordPress application password you connect is stored encrypted, as section 1 says.)
• No card details. They go to Dodo Payments on its own checkout page, and never to Passcite.

3. What it is used for

• Your account and sessions keep you signed in and keep your brands yours: every page and action checks that a brand belongs to the account asking. The IP address and user-agent on a session exist so a suspicious one can be recognised; they are not used to profile you or build an audience.
• A brand's profile, prompts and knowledge base are what its reports ask and what its articles are written from.
• The payment IDs and the ledger say which plan you are on, what you have left, and what each brand spent.
• A shared link's count tells you whether what you sent was opened, and when.
• A prospect's published details — the facts from its site and the map — help you decide whom to pitch and where to write; Passcite itself never contacts the business.
• A cited page's published contact tells you whom to write to about it; Passcite itself never emails that site or posts anywhere.
• The counters stop one address, one inbox or one account from flooding a door, and keep only a count until their window ends.

4. Email

Email is sent through Resend. A sign-in code goes to the address it was asked for, because proving that address is its whole job. To your account's address: billing — when a plan or its trial starts, and when a payment does not go through — and delivery — when a client report edition is waiting for your review or has gone out, and when a scheduled report could not run.

On the Agency plans, a published client report edition is emailed to the addresses you added to it. It is sent in your agency's name from our sending address, with replies going to your account's address, and it is the only email those addresses receive from us.

A pitch's email is never sent by us: it opens in your own mail app or Gmail, or you copy it, and it goes from your address.

There is no newsletter, and your address is not shared or sold.

5. Who else touches the data

• Vercel — hosting and serving this site, counting page visits on it, and passing every request to an AI model through its AI Gateway (United States).
• Neon — the Postgres database where the above is stored (AWS, US East).
• Google — sign-in, the favicon endpoint that serves site icons, YouTube's oEmbed endpoint for a cited video's channel, and Search Console and Google Analytics for a brand you connect, as described above.
• Resend — delivers the emails described above (United States).
• Upstash — the Redis store holding the counters described above, and short-lived caches: a market's map listings and what a business's site publishes, homepage pictures, the suggested prompts written in a market's language, and when a brand's sitemap was last read.
• Dodo Payments — takes the card and the payment for a plan or a top-up, as merchant of record, and receives what is listed under “When you pay”.
• Cloudflare — DNS, and email routing for hello@passcite.com.
• SerpApi — fetches the businesses Google Maps lists for a market you scan, Google AI Mode's answer to a prompt, and, for a brand with no Search Console connected, whether Google lists an article's page when its Proof asks again; it is sent the kind of business and the place, the prompt and the place it is about, or the page's address (United States).
• Google PageSpeed Insights — opens a prospect's homepage to picture it and time it on a phone, and a page you check in Site health to score it; it is sent that address.

A brand's prompts are sent to the engines that answer them — ChatGPT through the OpenAI API, Claude through Anthropic or a cloud that hosts Anthropic's models, Gemini through Google, Perplexity through Perplexity's API and Grok through xAI's, and Google AI Mode through SerpApi — as the questions themselves. To read a site, suggest tasks, read Brand Audit's answers, and write and check articles, Anthropic's models are sent the text of the pages read, the brand's profile and knowledge base, the answers being read, and the instructions you give; to draft a pitch's email, the pitch's figures and your agency's name; to draft an outreach email or a community reply, the cited page's or thread's text and the brand's facts; for a research, the keyword or the page's text and the brand's market. All of it but Google AI Mode goes through Vercel's AI Gateway. Your email address is never sent to any of them.

An assistant you connect (above) is sent what you ask it to read from your account — brands, prompts and answers, sources, tasks, articles, Proof, client reports, prospects and outreach — and that goes to the company that runs it, under its terms, not ours. It is never sent your Search Console or Google Analytics data, or what Indexing keeps.

6. How your data is protected

Security procedures are in place to keep your data, and the Google user data you let Passcite read, confidential:

• Encrypted in transit. Passcite is served only over HTTPS, and tells browsers never to use a plain connection to it (HSTS). What Passcite sends to Google's APIs, to its database and to the services in section 5 travels over connections encrypted with TLS.
• Encrypted at rest. The database is hosted by Neon, which encrypts everything it stores with AES-256. The credentials a connection gives Passcite — the Search Console and Google Analytics refresh tokens, a WordPress application password, a Webflow API token — are encrypted again by Passcite before they are stored, with AES-256-GCM, under a key kept in the hosting environment rather than in the database. A copy of the database on its own does not reveal them.
• Only your account reaches your data. Every page and action that reads or changes a brand first checks that the brand belongs to the signed-in account, so one account cannot read another's. Sign-in sessions are cookies that scripts on a page cannot read and that are sent only over HTTPS; they expire 7 days after they were last used.
• The least access to Google data. Search Console and Google Analytics are each asked for a single read-only scope, in a consent of its own, and only when you choose to connect them; signing in never asks for them, and Passcite never changes anything in either. To read them, a short-lived access token is made from the stored refresh token, used for that request and not stored. Google Analytics figures are never stored. Of what Search Console returns, Passcite keeps only what section 1 lists: each page's index status for Indexing, whether Google had an article's page on the day its Proof asked again, the figures a client report edition was sent with, and the searches you track.
• People. Only the founder has access to Passcite's systems and its database. Google user data is not read by anyone unless you ask for help with it or the law requires it.
• Taking access back. Disconnecting Search Console or Google Analytics in a brand's Integrations deletes its stored token and property at once — and, for Search Console, what Indexing kept — and so does removing the brand or deleting your account. A Google account connected to several brands is a connection on each, and the others keep theirs. You can also remove Passcite's access from your Google Account at any time, at myaccount.google.com/connections; Passcite then cannot read either one again until you reconnect it. Disconnecting an assistant in Settings → MCP deletes your consent and its tokens at once, and its next call is refused.

7. How long it is kept

• Your account and everything under it — brands, prompts, answers, tasks, articles, schedules, client reports and their editions, shared links, what was read of cited pages, researches and site checks, the pages and threads you acted on with their drafts, connections and the credit ledger — is kept while the account exists, so reports can be compared over time. Deleting a brand's connection deletes its credentials at once.
• A Search Console or Google Analytics connection — the encrypted refresh token and the property it reads — is kept until you disconnect it, remove the brand or delete your account, and deleted at once then, with what Indexing kept from Search Console. Google Analytics figures are not kept at all. Of Search Console's, an article keeps whether Google had its page on its Proof day, and client report editions keep the figures they were sent with; both stay while the account does, and go with it or when you ask (section 8).
• What is cached about a market and its businesses — the map's list, what a business's site says, and the picture of its homepage — expires after 30 days, or a day for a site that did not answer. A business you add stays with your list until you remove it.
• Payments are Dodo's records as merchant of record, kept by Dodo as tax law requires, after anything of ours is deleted.
• Sessions stop working 7 days after they were last used. Signing out deletes that session record immediately.
• A connected assistant keeps its access until you disconnect it: its access token lasts 60 minutes, and its refresh token stops working 90 days after it was last used. Disconnecting deletes your consent and its tokens at once.
• Page visits are held by Vercel rather than by Passcite, and only as counts. The identifier that distinguishes one visitor from another is discarded after 24 hours, so there is nothing left to delete on request and nothing to hand back if you ask for a copy.

8. Getting your data, or getting it deleted

Email hello@passcite.com from the address on your account and ask. You can ask for a copy of what is stored about you, ask for it to be corrected, or ask for all of it to be deleted. Deleting an account removes its brands and everything under them with it.

Being honest about the mechanics: there is no self-service delete button yet. Requests are handled by hand, and the aim is within 30 days. If that ever stops being fast enough, the button gets built.

9. Children

Passcite is a tool for businesses and marketing agencies and is not directed at anyone under 16. If you believe a child has given us personal data, email hello@passcite.com and it will be deleted.

10. Changes

If this page changes in a way that affects what is collected or who it goes to, the date at the top changes and the change is described here. The current version is the one you are reading.

The version of 12 October 2026 is a rewrite for the product as rebuilt on 10 October: an account before the first report, brands with their prompts and answers, tasks and articles, plans and top-ups paid in credits, and each WordPress, Webflow or Search Console connection belonging to a brand. The anonymous report request, the record of who opened a report, the cookie a request set and the private client links went with the code that made them, and are no longer collected. Later the same day, shared links came back in a new form — with an open count in place of a record of each visit — together with scheduled reports, client reports emailed to the addresses you add, and White label; and then Prospects: markets, the businesses the map lists in them, what a business publishes on its own site, and pitches with their links; and then Sources and Explorer: what the pages AI answers cite say, research, site health checks, and Google Analytics, connected with its own read-only consent; and then Outreach: the pages and threads you act on, the contact a site publishes, and the emails and replies drafted for you to send.

The version of 14 October 2026 adds section 6, on how data is protected, and says plainly what was already the case for Search Console: a client report's edition keeps its figures, and a search you track becomes a prompt. Nothing new is collected.

The version of 16 October 2026 adds the agency directory: what it keeps about each agency it lists, and how an entry is corrected or removed. It also names the pages about the businesses you prospect among those whose visits are never counted, which they already were; lists the daily limit on reading a brand's site to fill in its profile; and corrects how long a session lasts — 7 days from when it was last used, not from when it began.

The version of 17 October 2026 says what the agency directory keeps now that each agency has a page of its own — its offices, industries, the year it started and its contact page, all from its own site — that its pages show each agency's site icon from Google's favicon endpoint, and what happens to a request to be listed: kept until it is answered, then deleted, and limited to 5 an hour per IP address.

The version of 20 October 2026 covers the changes of 19 and 20 October: signing in with a six-digit code emailed to you in place of a link, the code kept only as a hash, and tries at it counted per inbox; a second cookie, kept 30 days, remembering which way you last signed in; Indexing, which keeps each page's index status from Search Console; an article's note of whether Google had its page on its Proof day, for which a brand without Search Console sends the page's address to SerpApi; a search you research being written into questions by Anthropic's model; removing a brand deleting its connections; and what the agency directory keeps to ask an agency whether its entry is right.

The version of 21 October 2026 adds connecting an AI assistant (MCP): what is kept to let it in, that what it reads goes to the company that runs it, that Google data is never sent to it, and that disconnecting it deletes its access at once.

11. Contact

Passcite, hello@passcite.com. It is the same address for privacy questions, support and everything else. See also the Terms.