Every change, dated and versioned.

If you sign up and never connect your AI, you now get a short email with the steps, instead of being left to work it out.

Connecting now defaults to Claude Code, with the setup prompt generated for you. The Claude app path is gone: it could never finish without a terminal.

The button that generates your connector prompt now fills the width of the dialog, so it is harder to miss.

Signing up no longer asks you to create a company first. Your workspace is made for you, so you reach the connect step straight away.

Signing up and signing in are now separate pages, so a new account is never one wrong click away from a login form.

The first screen after signup took almost six seconds to appear. It now loads immediately.

One-click MCP install for supported clients, instead of copying configuration by hand.

The connect step now waits for a real tool call from your AI before it counts you as connected, so you can no longer be told you are set up when you are not.

Starting a subscription and then leaving onboarding no longer loses the subscription.

Signing in counts as consent to session recording, and the privacy policy now says so plainly.

The connect step is now a simple 1 to 3 guide, with a way past it if you cannot connect right now.

A banner on your dashboard while your AI is not connected, so the next step stays in front of you.

An article that fails its checks now gives you a fix prompt you can copy straight into your AI.

Quality checks now catch an article restating the same point in different words.

Keyword research no longer gives up on a temporary provider error, and a failed search is not cached as though it were a result.

The homepage and landing page have been rewritten in plain language, with a clearer walkthrough of how the product actually works.

After connecting your AI you now land on your dashboard rather than the billing page.

Two new guides on the blog: writing SEO content with AI, and choosing an SEO automation tool.

The site now serves clean markdown to AI crawlers, so assistants quoting us get the real text rather than a page of markup.

Every quality check now shows where its numbers came from, so you can see what was measured and what was not.

Keyword data that could not be measured is now reported as unknown, rather than as zero.

get_search_performance can now show how a page is performing, not just a single keyword.

A mobile pass across the homepage, blog and onboarding.

If you'd already researched keywords for your project and then ran out of free lookups, your AI now knows it can still write with what you've already got instead of telling you the session is over.

New list_keywords tool: your AI can check which keywords it's already researched for your project for free, before spending a lookup asking again.

Asking for the same keyword twice in one request no longer uses up two lookups.

Your billing usage summary is simpler now: it shows what you've used without extra internal accounting detail you don't need to see.

You can now refresh your connected Search Console properties with one click, so a site you've just verified in Google shows up straight away, no page reload needed.

Uploading a company logo now actually finishes and saves. A storage permissions bug meant every upload failed after passing the permission check, so v0.14's logo-upload fix could pass its own gate and still never succeed.

You can now upload your own company logo. It previously only worked for a platform admin acting on your behalf, so an ordinary account owner could never set it themselves.

Admins managing more than one company now see the correct seat count and capacity for whichever company they're viewing, instead of their own company's numbers.

A project with no URL base set of its own could end up with another business's URL structure saved into its articles. An article like this is now held back and marked for review instead, with a note telling your AI what to fix.

Loading states and empty lists (projects, articles, team) now look and read the same way everywhere in the dashboard.

Dates across the dashboard, blog and changelog now always show in UK format (18 Jul 2026), instead of sometimes following your browser's locale or, on your billing page, always showing the US format.

A newly created business now appears straight away in your company list, instead of only showing up after a full page reload.

Every quality check now works from the same reading of your article, so the numbers behind your score agree with each other and with what a reader actually sees.

Your score now counts only the checks that apply to the kind of article you wrote, so scores are comparable across article types. Most scores shift by a point or two as a result.

Readability and keyword density are measured on the words a reader sees, with heading markers and link addresses left out of the totals.

The opening-sentence check now reads your whole first sentence, including openers that begin with a question.

How-to articles are now held to the how-to checks automatically, including structured data for the steps, with nothing to declare by hand.

Grading a draft and saving it run through exactly the same checks, so you get the same verdict either way.

Your AI now states the content type in the article itself, so the right checks apply from the first draft.

Your usage period stays anchored to your renewal date, even when the same renewal notice arrives more than once.

Lookups from a search that fails partway are returned to your allowance.

A custom URL set when you save an article is kept when that article is checked again.

Your usage period is always anchored to your renewal date, so everything you do counts towards the right period.

Annual plans show in full on the billing page, with the plan name, price and features for the yearly price as well as the monthly one.

Plan allowances are applied only from a recognised current price, so a retired price can never grant the wrong allowance.

Permissions on job records are now enforced at the database level as well as in the app, so a record can only ever belong to the account that created it.

The connector token page works for accounts that have not finished setting up a company.

Your AI now has an accurate description of when an article counts against your allowance, so it can tell you where you stand.

Your articles are checked against the URL they are actually published at, so your own URL structure passes cleanly.

A custom canonical URL set when you save an article is used by the quality checks as well as stored on the article.

This changelog. Every shipped change, dated and versioned, linked from the footer and the roadmap.

Your dashboard now tells you when something has shipped, once per release, with a link through to what changed. Dismiss it and it stays gone until the next version.

Trimmed the dashboard and the article page: the projects card and a redundant back link are gone, so the things you actually act on are easier to find.

Every article now shows its strict-judge verdict on the dashboard: nine criteria, each scored out of 100 against its own pass mark, with the judge's reasoning and the evidence it quoted.

New search_serp tool: live Google results for any query in your market, so your AI can find a real, citable source instead of inventing a statistic. One lookup per call.

The dashboard and connect page now show the exact order your AI calls the tools in, so you can ask for the right step to be repeated when something needs another pass.

Articles cannot be briefed until the business profile has positioning, audience and voice filled in. Free to fill, and it is the difference between content that sounds like you and content that sounds like anyone.

Discussion mining now searches your project's country. It previously searched the US for everyone, so non-US profiles were collecting the wrong buyers' words.

Articles keep their researched volume and difficulty on the dashboard, and the score column now says what it is: the deterministic gate score, separate from the judge's verdict.

A strict judge now stands between a draft and publication. Deterministic gates check structure; the judge checks whether the writing is actually good, quoting the article as evidence, and the server verifies those quotes exist.

Articles are charged only when they pass the gates, never for briefs or failed attempts, and re-grading the same article is free while you iterate.

Lookup and article allowances are now reserved atomically, so parallel requests can no longer slip past the limit or double-charge.

Keyword research runs against the country and language your business sells into. Volumes, difficulty and search results are market-specific, and an unset market is disclosed rather than silently defaulting to the US.

Every MCP tool call is logged, so usage questions have answers instead of guesses.

Free-tier limits are enforced server-side: ten lifetime lookups, one graded article, honest refusal messages when they run out.

Done-for-you backlink packages, ordered straight from the dashboard.

Connecting your AI is now one copied prompt per client (Claude, Cursor, OpenAI, Gemini), and onboarding waits for a real tool call before moving on.

Business profiles are capped per plan, enforced at creation rather than politely suggested.

Connect Google Search Console and the writing loop reads live performance: current position, clicks and impressions for the keyword each article targets, so re-briefs aim at what is actually underperforming.

Blog index with topic filters, search and grouped sections.

Dropped the bring-your-own-keys tier. Managed keys only, one less thing to configure and one less way to be misconfigured.

A free tier: connect your AI, profile your business and grade your first article without a card.

Existing site content is scanned for internal-link targets, so articles link to pages that really exist.

Writing now requires research first: the AI must show you real volume and difficulty and let you pick the keyword before it gets a brief.