Updates
Changelog
Product changes and what they mean for your workflow.
These notes describe implemented changes in the codebase. Dates are the recorded commit dates; availability in your workspace depends on deployment, access and configuration.
November 11, 2026: Work on a page with the agent
A chat about one of your content pages now opens as a workspace: the page beside the agent’s suggestions, its outline and its SEO fields and checks, with the chat underneath. Apply, dismiss or revert each suggestion, or apply every ready one at once; highlighted text shows where each one lands. Suggestions that need a fact from you, and any text you select, can be quoted into the chat without sending anything. You can also edit the page yourself, export it as Markdown, HTML or plain text, and mark it implemented from the same screen. Work with the agent on an Action for one of your content pages opens this workspace.
For your team: open an Action on one of your articles or guides and choose Work with the agent, or use Analyze a page in Actions.
November 10, 2026: Content optimization
After each visibility run and crawl, CiteLadder groups the competitor and publisher content pages (articles, guides, comparisons, reviews and the like, never product, category or home pages) that answers to your tracked prompts cite into content angles by topic, and matches each angle to at most one of your crawled content pages. An angle cited often enough across enough sites becomes an Action in Actions, with the rest of your work: on your matching page when that page is not cited for it, or on a planned page when no page matches. An Action’s Why this matters lists the cited pages and prompts behind it. Write this page and Work on this open the Agent with the matching workflow. Finding the angles uses no credits.
For your team: run a visibility audit and a Site Health crawl, then open Actions to see the angles competitors and publishers are cited for and you are not.
November 3, 2026: A faster app and quicker sign-in
Each screen now loads its data in one round of requests instead of several, opens without waiting on code for other screens, and shows one loading state rather than several in turn. Refreshing keeps the data you were reading in place. Signing in and out is faster and keeps the page drawn while it happens.
For your team: nothing to change; screens and tables should open and settle sooner.
November 2, 2026: Safer sign-up
Sign-in and sign-up now carry the Terms of Service checkbox on the form itself, so there is no separate review screen after you sign in. Sign-up, verification-email and password-reset requests are protected by a Cloudflare Turnstile check, and sign-up refuses disposable email domains. Signing in has no Turnstile check. Sign-in and sign-up also have a new full-page layout.
For your team: invite colleagues with their work address; disposable inboxes cannot create an account.
November 2, 2026: Faster, more reliable audits
Audits now run several answers at once instead of one at a time, and a Site Health crawl can no longer hold up a visibility run. ChatGPT API answers use a faster OpenAI model and one search round per answer, as the ChatGPT app does. An answer that comes back with no text is now a failed answer and is retried, never counted as an answer that did not mention you. When the Google AI Overviews provider pauses or refuses the account, the rest of that engine’s answers stop with a clear reason, and run progress counts failed answers as finished.
For your team: expect runs to finish sooner. Brand mention rates can rise slightly where empty answers used to count as “not mentioned”; compare runs from after this change.
November 2, 2026: Action priority reads High, Medium or Low
Actions now show priority as High, Medium or Low instead of a raw score, on the list, the detail page and New chat. Priority comes from the Action’s strongest finding, and the “evidence systems” count is gone because it did not measure independent evidence; the detail page names the source instead, such as Site Health. Long Action titles stay on one line, with the full title on hover.
For your team: sort by priority and start with the High items; the order may differ from before.
November 2, 2026: Site Health issues you can act on
A page’s detail view now leads with the issues on that page, worst first, each with its evidence and fix, followed by advisories, scores and a collapsed Technical details section. “Defects” are now called issues. A broken or redirecting link that appears across most of the site, such as a footer link, is reported once with the pages that carry it, instead of once per page. A new check names sections that have internal links but cannot be reached from the homepage, and a sitemap listing a page you chose to noindex is now advisory.
Crawls also reach their page limit when some URLs turn out not to be pages, follow links through repaired tracking URLs, and stop counting non-HTML links as failures. The Overview only reports a scored-cohort change when a page kind joined or left or its share moved materially, and says so once.
For your team: run a new crawl to see the regrouped issues; counts can drop sharply where one sitewide link used to raise an issue on every page.
November 1, 2026: The Prompt Builder
Generating prompts now happens in the Prompt Builder, a four-step sheet on Prompts that replaces the Generate dialog and Build with Agent:
- About your business: confirm what you sell, who buys it, the situations and attributes that shape the choice, and more, plus a note for this run only.
- Topics: keep, rename or add topics, or ask for buyer-facing suggestions built from your offerings.
- Distribution: choose the count, branded share, competitor-alternative share, intent mix, personas and whether to use your Search Console or keyword-research searches, with each search shown so you can untick it.
- Review: accept suggestions with a coverage line per topic and per share, and generate more where one comes up short.
Suggestions are now short buyer questions about one need each, capped at 16 words and planned across awareness, consideration and decision. Near-duplicates and questions joining two alternatives are dropped, topics get questions in proportion to the offerings they cover, and a branded question may only describe you with attributes you confirmed. Each candidate is also checked for whether a good answer would have to name a business.
Agent → Context shows the business map read-only with Edit in Prompt Builder, and Brand facts is its own tab. The Agent’s prompt discovery skill now writes a prompt plan following the builder’s steps and links to it, without writing prompts. A connection with Allow changes can follow the same steps over MCP with preview_topics, confirm_topics, read_seed_queries and generate_prompts; suggestions it creates wait in Prompts for you to accept.
For your team: open the Prompt Builder from Prompts, check the first step’s business facts, and confirm topics before generating your next set.
Build prompts with the Prompt Builder
November 1, 2026: A branded Overview PDF
The Overview PDF now opens with a CiteLadder cover, a strip of headline figures with what each measures, the next action with its target, and a plain-English “How to read this report” in place of internal codes. Engines appear by name, dates read naturally, missing values print as a dash rather than “Unknown”, and change figures appear only once a comparable earlier run exists.
For your team: share the PDF with stakeholders as it is; it no longer needs editing to remove IDs or jargon.
October 18, 2026: Remove competitors and delete topics through MCP
A connection with Allow changes can now stop tracking a competitor and delete a topic, both by name; a deleted topic’s prompts stay tracked without a topic. Every MCP tool parameter now describes what it takes and where its value comes from, so assistants pick the right tool and arguments the first time.
For your team: ask your assistant to tidy your competitor list or topics directly; it asks before each change.
October 12, 2026: A faster Agent that remembers the chat
The Agent now reads several pieces of data at once instead of one at a time, and shows each read as it runs. A follow-up builds on what it already read and wrote earlier in the chat instead of starting over. While it works, you can open a short summary of its thinking, and the reply and document fill in as they are written. A saved reply’s run summary counts the model calls and data reads it used.
For your team: ask broader questions in one message, and keep refining in the same chat; the Agent carries the earlier work forward.
October 11, 2026: Confirmed changes through MCP
When you connect an assistant, you can now tick Allow changes. The assistant can then add and edit topics and prompts, archive prompts, add competitors, launch and cancel audits, create audit schedules and update or declare Actions, in workspaces where you are a Member or above. Larger changes show you a preview first and happen only after you agree; added prompts are active at once. Existing connections stay read-only.
For your team: to let an assistant keep your prompts and audits up to date, revoke its connection and connect again with Allow changes ticked.
October 10, 2026: Markets
Measure the same prompts from more than one country and language. Growth plans can add 3 markets to their projects and Scale plans 10, in the project edit panel. Each market runs as its own audit, launched or scheduled together, and Visibility reads one market at a time with a By market table comparing their latest runs. Engines that cannot measure a market are skipped there or refused before anything runs, never counted as failures.
For your team: see where your brand is named in each country you sell in, without mixing one market’s answers into another’s.
October 10, 2026: Fact-checking pilot
Workspaces in the fact-checking pilot can confirm brand facts (pricing, plans, integrations, markets, policies and more) in Agent → Context. Audits then check the factual claims answers make about you against those facts, and the new Accuracy tab in Visibility shows how many claims were supported or contradicted, each contradiction quoted next to your fact. Assistants connected with MCP, and the in-app Agent, read the same results with read_fact_checks.
For your team: catch answers that quote an old price or a missing integration, and fix the pages behind them.
October 10, 2026: MCP moved to api.citeladder.com
The hosted MCP server is now at https://api.citeladder.com/mcp, beside the REST API. The old https://citeladder.com/mcp address no longer answers.
For your team: remove the old CiteLadder connector from your assistant and connect again with the new address, or use Connect in CiteLadder.
October 10, 2026: REST API and API keys
Paid plans include a REST API at https://api.citeladder.com/v1. Owners and Admins create keys in Settings → API keys, choose what each key can do and which projects it reaches, and see each key’s last use. Keys read projects, prompts, audits, visibility, Site Health, AI Traffic, Performance and Search Intelligence, and, with the matching scope, manage prompts, topics, competitors, audit schedules and actions or launch an audit with a credit ceiling. A key never does more than its creator’s current role allows.
For your team: feed CiteLadder results into your own dashboards and keep prompts and competitors in step with your other systems, without sharing a login.
October 10, 2026: Prompt generation grounded in your search data
Generated prompts now borrow wording from how your buyers already search. When Search Console is connected or you have published keyword research, suggestions for a topic with matching searches are written with a few of your own non-branded searches for that topic as examples, and suggestions phrased this way carry an Informed by your search data tag. Nothing extra is fetched or charged, searches are never copied as prompts, and impressions or keyword volume are never shown as AI prompt volume. Without that data, generation is unchanged.
For your team: connect Search Console before generating, and expect suggestions that sound more like your real buyers.
Read about grounded generation
October 10, 2026: Google Cloud crawler logs
Connect Google Cloud load balancer and Cloud Run request logs to AI crawler
logs. A Cloud Logging sink sends crawler requests to a Pub/Sub subscription in
your project, and CiteLadder pulls it every few minutes as its own reader
account; no token or endpoint is involved. The connect dialog generates the
gcloud commands, including a label that proves the subscription is yours, and
Verify subscription activates the source. When CiteLadder recognizes new
crawlers, the source asks you to update the sink filter.
For your team: if your site runs behind a Google Cloud load balancer or on Cloud Run, you can now measure AI crawler visits with complete daily coverage.
October 10, 2026: Ads in ChatGPT answers
AI Visibility has an Ads view: how often ChatGPT Search answers showed paid ads, who advertised, your own ad share, the prompts that surface ads and each ad as text. Each ChatGPT Search answer in a run lists its ads apart from its citations. Ads never count as citations or change your visibility scores, and other surfaces read not applicable.
Assistants connected with MCP, and the in-app Agent, can read the same numbers with read_ai_ads.
For your team: see which competitors pay to appear in the answers your buyers read, and on which prompts.
October 10, 2026: AI crawler logs on paid plans, now with Amazon CloudFront
AI crawler logs are available on every paid plan. Connect Amazon CloudFront through an Amazon Data Firehose stream, alongside Cloudflare, custom webhooks and file uploads, with an optional filter Lambda that sends crawler requests only. Log endpoints moved to https://api.citeladder.com. A source that stops delivering now shows as Stalled with the reason.
For your team: connect the CDN you already run and watch the source row; a stalled source means that day’s coverage is not complete.
October 10, 2026: Answer perception
AI Visibility has a Perception view. Answers that name you or a competitor are read for sentiment: net sentiment from −100 to +100 with its coverage, the themes answers praise or criticise with quotes taken word for word from the answer, the sources cited alongside criticism and how often answers recommend you. Each answer in a run also shows how it portrayed every business it named. Classification uses no credits.
Assistants connected with MCP, and the in-app Agent, can read the same numbers with read_perception.
For your team: check what answers say about you, not only whether they mention you, and read the quotes before acting on a label.
October 10, 2026: Scheduled audits documented
The AI Visibility guide now explains how to schedule audits from Runs: repeat a prompt set one time, every few minutes, hourly, daily or weekly, see why a schedule paused and resume it. It also clarifies that answers are collected from the consumer apps and from provider APIs.
For your team: schedule the audits you rely on so your baseline keeps measuring without a manual launch.
October 10, 2026: Connect an assistant in one click
A single Connect button in CiteLadder adds the MCP connection to your assistant, and the approval page where you choose workspaces is redesigned. More clients can connect, including Cursor and browser-based assistants. MCP reads now match what the app shows for Actions, earned sources and audit results, and the in-app Agent reads evidence through the same catalogue.
Queued work such as crawls and audits also recovers more reliably after an interruption.
For your team: connect your assistant from the app, approve only the workspaces it needs, and ask it about Actions and AI traffic as well as visibility.
October 9, 2026: Earned sources, AI Shelf and AI Traffic
Earned sources. CiteLadder now opens one kind of earned Action: a page AI answers cite that lists your competitors and not you. It is measured on the tracked prompts that cite the page. The cited-page view leads with where you stand on that page.
AI Shelf. Product visibility no longer reads as 0% when every answer for a target failed; it shows as unavailable. Owned products are matched more strictly, shelf numbers refresh after each audit, and shelf findings link to Actions.
AI Traffic. The screen leads with AI referrals, preset windows now complete correctly, insights stay current, and the crawl-log Worker template batches requests so busy sites no longer lose data. Re-download the template if you installed it earlier.
For your team: use the earned brief for outreach, and treat any later change on the cited prompts as evidence to review, not proof of cause.
October 9, 2026: Keyword gaps become Actions
Search Intelligence adds a guided first run, a cancel control and clearer results. The cost estimate is now a ceiling the acquisition cannot exceed.
When a published competitor dataset shows a search a competitor ranks for and you do not, CiteLadder opens an Action for the page that covers it, or a planned page. After you ship the work, the Action checks for up to 90 days whether your site appears for that search.
For your team: review keyword gaps alongside your other Actions and declare the work implemented once the page is live.
October 8, 2026: Prompts, Actions and the Agent
Prompts and Visibility. A new project starts with no prompts: generate candidates or add your own on Prompts, then accept the ones worth measuring. “Visibility” now means the share of answers that name your brand everywhere it appears. A coverage strip under the headline shows how many answers and engines the numbers rest on, and each failed answer gives a plain reason and next step.
Actions. Marking an Action implemented now shows a measurement checklist with the go-live date. Each check is read by the evidence that can answer it, so Actions mixing page and search checks can finish, and a waiting page check offers Run crawl now.
Agent. Replies stream as they are written. You can stop a turn, try again, regenerate or edit your last request, and each reply lists the sources CiteLadder read for it.
Connected data and onboarding. Bing Webmaster Tools data now reaches Performance, and you can connect Search Console, Google Analytics or Bing from the screen that needs them. Onboarding preselects the competitors it finds, and category, buyer type and market can be edited after setup.
For your team: review generated prompts before measuring, and use the checklist to see what each implemented Action is waiting for.
October 8, 2026: Site Health scores what matters
Web and AEO scores now count only the checks that affect whether search and AI engines can reach, read and trust a page; other checks still raise issues without moving a score. Crawler access is reported for the whole site, coverage is shown by page type, and three unscored checks are added: content recency, entity profiles and sitemap and canonical agreement.
For your team: scores from earlier crawls may differ from later ones; compare crawls run after this change.
October 7, 2026: Faster crawls and a refreshed app
Site Health crawls read pages faster, and a crawl no longer stalls while follow-up work is waiting. The app adopts the same design as the public website.
For your team: run a new crawl to see the current website with the faster crawler.
September 26, 2026: Agent panel and theme switch
Dashboard screens gain an Agent panel: open it from the top bar to chat beside the screen you are viewing. Search Intelligence attaches the rows you are viewing, and a Site Health issue starts the panel with a question about it. The panel shares chats and outputs with the Agent workspace, and Open in Agent continues the same chat there.
The account menu’s theme checkbox is replaced by a one-click Light/Dark switch in the application and onboarding headers, and the dark theme moves to near-black neutrals.
For your team: ask about the screen you are on without leaving it, then move to the full workspace when the deliverable needs more room.
September 26, 2026: One Agent capability
The Agent now has one access setting, one model choice and one usage allowance for all of its work. This follows the move of content and growth work into the Agent workspace.
For your team: use the Agent as the common place to investigate evidence and create reviewable deliverables. If a control is unavailable, check the displayed access or funding reason.
September 25, 2026: Agent, Actions and reviewable outputs
A dedicated Agent workspace brings together new chats, Actions, Skills, Context and saved conversation history. Evidence handoffs let you begin with an existing finding instead of restating it.
Chats own a deliverable with revision history, user edits, copy and Markdown export. Long-form content requires outline approval before drafting. Runs expose distinct queued, running, cancelled, failed and limit-stop states.
For your team: choose a target, review its supporting evidence and refine the output before using it. The Agent does not publish or make external changes.
Work with outputs and revisions
September 25, 2026: From implementation to measurement
Actions group related findings around a target and carry the work’s lifecycle. Explicit implementation declarations can reference the exact Agent output revision that was shipped.
Later evidence checks the expected outcomes of that declaration. The result distinguishes implemented work, work still being measured and work whose latest observation verified the expected checks. Search Demand connects actionable signals with their associated Action.
For your team: generating an output is separate from shipping it. Declare implementation after making the change, then review what later measurements support.
September 24, 2026: Appearance and workflow refinements
The application and onboarding gained a device-local Light/Dark theme preference. Website page classification improved for product pages under category-style paths, alongside corrections to checkout, import and dialog behavior.
For your team: choose your preferred appearance; it is now a header switch (see September 26). For classification-dependent findings, inspect a later crawl’s page evidence rather than assuming old observations changed.
How to read these notes
A change in the product does not rewrite historical evidence. Audits, crawls, imported datasets and Agent revisions preserve the observations or work captured at the time.
When comparing before and after an update, keep measurement scope and availability in view. Contact support if your workspace’s behavior differs from the guide.