Changelog

The changelog opens with the ten most recent updates in full; everything earlier is listed below them as a single line with its date and type, and each one links to a page of its own. Those pages have a stable address you can share or bookmark, and the RSS feed now links to them. Links you already saved to an entry on the timeline still land on it. The page itself is a fraction of its former size, so it opens noticeably faster on a phone.

Scheduled AI-visibility scans and recurring audits now take turns across workspaces, so one workspace with many brands or targets — or one that has run out of credits — can no longer hold up everyone else's refresh. A cancelled deck stays cancelled even if a pipeline step was already running, and a deck whose credits were refunded can no longer be resumed: start a new one instead (resuming is now limited to editors). Password-reset emails are rate-limited per network and per address. A promo code can no longer be used twice by opening several checkouts before paying; if you already have a discounted checkout open, finish it or try again in 30 minutes. Subscription refunds now set the period's credits aside the moment you ask, and a workspace can keep up to 50 active recurring audits.

When you change your workspace's public address, widgets already embedded on your clients' sites at the old address keep working and keep sending leads to you. The old address is reserved for your workspace permanently: no other workspace can claim it, and if a workspace is deleted, its addresses stop resolving instead of passing to someone else.

Several places treated "created this" or "belongs to the agency" as if it meant "may manage it". Your agency's tax ID, VAT number, invoice address and phone are now readable only by owners and admins — at checkout and at the database itself. An admin can no longer remove an owner, or bring back a suspended owner through a fresh invitation. Someone removed from or suspended in your agency can no longer read its decks through an API key or the MCP tools, does not receive the weekly AI-visibility digest, and does not get your agency's details or live share links in their personal data export. And if you delete your account while you are an agency's only owner, ownership passes to an admin or member — never to a viewer — and cancelling the deletion takes it back.

When you first reach Pitchko through a campaign-tagged link, we keep that link's utm_source, utm_medium and utm_campaign labels in a first-party cookie and save them to your agency account as its signup source when you create it. Later visits never overwrite the first one, the labels do not identify you, and they are never sent to a third party. Our Privacy Policy and Cookie Policy describe it under a new "Signup source" heading.

A deep crawl could already tell you that 480 pages were missing an H1. It could not tell you that they were all the same page template, which is the part that turns a report into an afternoon of work instead of a quarter of it. The clustering that answers this had been written and tested for a while and was wired to nothing, so every crawl report shipped without it.

It runs now. A finished crawl groups the pages it read into templates — product pages, blog posts, the homepage — and a new Templates tab lists them with the URL shape each one matches and how many findings belong to it. The issues list groups itself the same way, so a finding that repeats across a template appears once, under the template, rather than forty times under forty addresses.

Every count on that tab is over the pages the crawl actually read, and the tab says so. A crawl visits a sample of a site and never learns the site's true size, so "480 pages sampled" is the honest figure and a site-wide total would be one we invented. Where the crawl was too large for us to read every page back, you still get the templates, and the per-finding counts stay the ones measured across the whole crawl rather than quietly shrinking to the part we re-read.

The tab appears on a deep-crawl report and on a report you share with a client. An instant single-page audit has nothing to cluster, so it does not show one.

If you connect Claude, Cursor or another MCP client to Pitchko, it now offers three ready-made prompts next to the deck tools:

• Summarise a deck — who the deck is for, the findings it rests on with their figures, and the actions it recommends, in order.
• Compare with competitors — where the brand leads or trails each rival the deck names, with the deck's own evidence.
• Write the client's next steps — a short follow-up message with concrete next steps, leaving the client and agency names for you to fill in.

Pick one, give it a deck id, and optionally a language (English or Turkish). Each prompt reads the deck through get_deck_content, keeps the deck's numbers as the deck's claims, and never adds figures of its own. They appear for any API key with the decks scope and cost nothing to run on our side.

The Pitchko MCP server has a new read-only tool, get_deck_content. Until now an assistant connected through MCP could list your decks and fetch their download links, but it could not see what a deck actually says. It can now read a completed deck's generated text — every section in slide order, in the deck's own language — so you can ask it to summarise the strategy a deck proposes, compare two decks, or point out what one leaves out. It uses the same decks scope as the other deck tools and the same access rules: a key only reads decks its user can already open. Nothing about cost, prompts or the underlying research leaves Pitchko with it, and like every Pitchko MCP tool it cannot create, change or spend anything.

If you opened Billing → Plan since the move to credits, it told you that you had 0 decks available — one click away from a buy button. You didn't. The line was still reading the old deck counter, which was converted into credits and emptied when everything moved onto one wallet. Nothing had been spent; only the reading was wrong.

The plan page now shows your credit balance, and — when your plan includes credits — how much of it came with the plan and how much you bought. The two are shown apart because they behave differently: plan credits roll over up to a cap, purchased credits never expire and can be refunded while unspent.

Two related fixes:

• The price on the plan card is the one you pay. The renewal line now reads the plan you actually bought. It could show a retired price instead; if your account was set up by us rather than bought through checkout, no charge is shown at all, because none will be made.
• A heads-up before you run dry. When your balance can no longer cover a full deck, the overview now says so at the top, with a link to top up. It stays quiet while you can still make one.

No credits were lost and no balance needs correcting.

New passwords now need at least 8 characters, when you sign up and when you reset a forgotten one. Nothing changes for an account you already have: if your current password is shorter, you still sign in with it exactly as before, and the new rule applies the next time you choose one. Separately, the cookie that credits a partner when you arrive through their referral link is now set only after you allow marketing cookies. If you decline, it is not set and any earlier one is removed — which means that partner is not credited for your sign-up. The cookie policy lists it under Marketing now, and it also stops calling the light/dark theme choice a cookie: that preference lives in your browser's local storage and is never sent to us. Finally, agency and client logos can now be uploaded as PNG, JPG or WebP; new SVG uploads are no longer accepted, because an SVG file can carry code. Logos you already uploaded as SVG keep working. And the logo you pick while setting up your agency is now actually saved — until now it showed in the setup wizard but the agency was created without it.