Changelog
What's new in Ghost Agent Lab
Every new feature and improvement, newest first. The same updates appear under “Updates” in the app.
October 2026
-
A tidier Account page
Account now opens with who you are and a short summary of how well your account is protected: whether your email is confirmed, whether two-step sign-in is on, and how many devices you're signed in on. Links under it jump to each section, and an unconfirmed email gets a clear prompt with the button to send the link.
-
A clearer email confirmation page
The page your confirmation email opens now names the address you're confirming and lists what confirming turns on: email alerts and reports, inviting teammates, and getting back in if you forget your password. If you open an old link after confirming, it says you're already confirmed instead of calling the link expired, and an expired link takes you straight to where you can send a new one.
-
The free AgentScore is for your own website
The free AgentScore scan on ghostagentlab.com now asks for an email address at the same domain as the website, like you@example.com for example.com, so each report goes to someone at that company. To scan other sites, such as clients' or competitors', add them in the app. Free reports also only compare a score with our study of online stores when the site is a store.
-
More small fixes
The "Live" badge at the top of each site turns into "Updated … ago" once traffic stops, without reloading the page. On the Fixes page, the Owner list only offers people who can open that site. And the monthly PDF no longer shows a peak of 1 a day for a month with no AI requests.
-
Smaller fixes across the app
People with view-only access can now check an IP address on the Verification page. When every competitor scan fails, the comparison says so instead of waiting for ever. And "Since your last visit" on the home page no longer resets half an hour into a long session.
-
Stay signed in while you use the app
If you use Ghost Agent Lab regularly, you now stay signed in. Before, your browser signed you out 30 days after you signed in, however often you'd used the app since.
-
Confirm it's you before changing your email or adding two-step sign-in
Changing your email address or setting up two-step sign-in now asks for your password, plus a code if two-step sign-in is already on. If you only sign in with Google, sign in again first; the change then works for 15 minutes. Someone using a computer you left signed in can no longer take over your account this way.
-
Each agent's page shows your access policy
An agent's details now show whether your access policy allows, limits or blocks it, with a link to change it.
-
AgentScore reads more sites correctly
Sites that only work with www. in front are now scanned instead of reported as unreachable. Compressed sitemaps count, and an out-of-date sitemap in robots.txt no longer hides a good one at /sitemap.xml. A robots.txt that names an agent with a version, like "GPTBot/1.1", now shows as blocking it on Access policy and in alerts. And a free scan of one page is no longer shown for a different page on the same site.
-
Choosing a new password is clearer
The page a reset link opens says which account it's for, tells you as you type whether the password is long enough and whether the two match, and saves to your password manager under the right email. Afterwards you see that it worked.
-
Free reports show what to fix first
A finished free AgentScore report now picks the three findings to fix first and says how Ghost Agent Lab keeps each one fixed. Our website also has new platform guides for Shopify, WooCommerce, BigCommerce, Adobe Commerce, Webflow and WordPress, and topic pages that gather everything on bot protection, checkout and more.
-
A clearer password reset
After you ask for a reset link, the page says what to do next and lets you send the email again or use a different address. If you usually sign in with Google, a reset link also adds a password.
-
A plan calculator, and a live AgentScore benchmark
Our pricing page now has a plan calculator: enter your sites, team, AI agent traffic and the journeys you want tested, and it picks the plan that costs least, extra usage included. And the benchmarks page now shows how every site AgentScore scanned in the last 90 days is doing, updated every Monday. Only totals are shown; no site is ever named.
-
Add all your client sites at once
Agencies and teams with several brands can now paste every site in one go on Add several sites, with the email of each site's developer. Each developer gets the setup steps for their own site, and readiness scans start on their own. On All sites, every site with something in the way of its setup is marked in the Sites table, not just the first three.
-
More free tools: see your site as ChatGPT does
Three more free tools on our website. "What does ChatGPT see?" shows any page the way AI assistants fetch it, without JavaScript, next to what your visitors see. The llms.txt generator drafts an llms.txt from your home page. And the AI crawler checklist is a one-page PDF of which AI agents to allow and what to check.
-
Our own visits no longer count as your traffic
Readiness scans that test your bot protection as ChatGPT, Claude, Perplexity and Googlebot, and Ghost Agent test runs, are left out of your traffic, so they can't raise impostor or "being blocked" alerts. On Akamai and CloudFront, add the query-string field shown in the updated setup steps.
-
Fairer "vs the week before" changes
Changes now compare with the period before up to the same time of day, so a day that's only partly over no longer makes your AI traffic look like it fell, including in the Monday report.
-
AgentScore reads more prices correctly
WooCommerce product prices, and prices written like $12.990 or R$ 1.299,90, are now read correctly, so they no longer cause false "no price" or "price mismatch" findings.
-
Switching sites always shows that site's settings
Access policy, llms.txt and Agent traffic now show the site you switch to, so a change you save goes to that site. Agent traffic no longer fails to load when several requests arrive in the same second.
-
Verification loads faster
On busy sites, Verification now opens in under a second, even for 30 days. The numbers are the same as before.
-
European prices read correctly in AgentScore
Prices written with a dot for thousands, like "€ 1.299,-", were read as €1.30, so the price check could fail and cap your score at 79 even though your prices agreed. Your next scan reads them as €1,299.
-
A clearer sign-up page
Sign-up shows what your free trial includes, tells you straight away if an email address can't start a trial, and counts down to the password's minimum length.
-
Ask Ghostie why your data isn't showing yet
Ask Ghostie knows what's in the way of each site's setup, the same as “Something's in the way” on All sites: a scan that couldn't load your site, logs that haven't arrived or arrive with the wrong times, or a setup link nobody opened. Ask “Why don't I see any data?” and it names the problem and the page that fixes it.
-
Free tools: check a bot and test your robots.txt
Two free tools on our website. "Is this bot real?" checks whether a visitor claiming to be GPTBot, Googlebot or another AI agent really is, using the same checks as your agent traffic. The robots.txt tester shows which AI agents your robots.txt lets in, and the rule behind each. We've also published how AgentScore is scored: every check, its weight and what caps the score.
-
A cleaner sign-in page
Sign in, sign up and password reset have a new look, with "Forgot password?" right next to the password field. They work better on phones and in dark mode too.
-
Joining a colleague's team doesn't need a card
If you create an account from a team's join link, you go straight into that team once they let you in, instead of being asked for a card. While you wait, the Plan and usage page tells you who you've asked to join.
-
Scans no longer wait behind Monday's visibility checks
Free and app scans, Ghost Agent runs and alerts now start ahead of the weekly AI visibility checks.
-
Running scans update faster
Pages that show a running scan or Ghost Agent run update faster, and pause while the tab is in the background. Scans also load store pages faster.
-
Bot challenges recognised on every page
AgentScore now recognises the same bot challenge pages on product, cart, checkout and sampled pages as on your home page.
-
Ghost Agents see past long menus
Ghost Agents now read the part of the page they're looking at first, so a long menu or filter sidebar no longer hides the products below it, and they can scroll to the product grid and add items to the cart.
-
Sample data no longer makes agents look new
Overview, All sites, Mission Control and Ask Ghostie only list an agent as new after its first real visit to your sites. AI-assistant referral visits now follow your plan's traffic history, like the rest of your traffic data.
-
An invited teammate who joins another way takes one seat
If someone you invited joins through your join link or company domain instead, their invitation is closed and they take one seat, not two.
-
The same alerts on every page
All sites, Mission Control, Ask Ghostie and your reports now list the same alerts as each site's Alerts page, including paused scheduled checks and log sources that have gone quiet.
-
A lighter Mission Control
Mission Control covers up to 25 sites and says so when you have more, and All sites and Mission Control find new agents without reading your whole history; live figures still update every 30 seconds. On Agent traffic, “When agents last read your top pages” now covers the period you're viewing, at least 30 days.
-
AgentScore no longer invents a pricing page
Sites that answer every address, by sending unknown pages to the home page or serving the same app everywhere, were sometimes scored as a software site with a pricing page or cart that doesn't exist. They aren't any more.
-
Ghostie's numbers match the dates you're looking at
Ask Ghostie now answers for the date range on the page you're on. If it can't answer because of a problem on our side, it says so, and the question doesn't count towards your daily limit.
-
See the latest failed journey at a glance
All sites now shows the latest failed Ghost Agent check as a strip of screenshots, with the step where the agent got stuck outlined in red and what it was thinking there.
-
Your sites live in the top bar
The list of sites has moved out of the sidebar, leaving it for the sections of the app. To switch site, or see each site's AgentScore at a glance, open the site menu at the top left of every page. It marks “No data” beside a site whose agent traffic hasn't arrived yet (amber) or has stopped for more than a day (red). All sites still compares them side by side.
-
Agent traffic opens in a moment, even on busy sites
Agent traffic, Overview, the agent directory and search now read from running totals kept as each request arrives, instead of counting every request each time you open them. On a site with millions of agent requests, most sections now load in a few hundredths of a second. The numbers are the same as before.
-
Setup that starts from your question
When you add your first site, say what you want to know first: whether AI assistants can find you, whether an AI agent can buy from you, or who's scraping you. A scan or a checkout test starts straight away, and your setup checklist leads with the steps that answer it. All sites also explains when setup has stalled, like a setup link nobody opened or logs with the wrong times, and what usually fixes it.
-
Let a finance colleague handle billing
Owners can now make anyone on the team a billing manager, even a viewer. Billing managers can change the plan, update the card and download invoices, without being able to change sites or the team.
-
Find and review teammates
Search and filter the team by role or status, see every role's permissions side by side, and open anyone's page to see how they joined, their sites, alert emails and recent activity, or sign them out everywhere.
-
Change your email, download your data or delete your account
Change the address you sign in with (we confirm the new one first), download a copy of what we store about you, or delete your account, all from Account settings.
-
A failed AI agent task caps the free scan's score
When the free scan's AI agent tries a task, like adding a product to the cart, and can't finish it, the score now can't go above 59, however well the rest of the site does. A slow finish is still a warning, and a run that stops for reasons that aren't your site's still doesn't count.
-
API keys stop when the person who made them leaves
An API key now works only while the person who created it is an owner or admin here. If they leave, lose that role or are suspended, their keys stop, so nobody who has left keeps access. Keys always read every site, and the key list shows who created each one.
-
Alerts and reports keep trying when Slack or email is down
During a brief Slack or email outage, alerts and weekly and monthly reports now keep trying for about an hour instead of giving up after one retry. A report that reached only some people is retried only for the rest, so nobody gets it twice.
-
Ghost Agents are even more careful at checkout
Ghost Agents already refuse to press “Place order” or “Pay now”. Now they also won't press Enter where it would place the order, even behind an icon-only button, and they double-click only to select text, since a second click can't be checked first. Search boxes still work with Enter.
-
Agent traffic and Overview load faster
On sites with a lot of agent traffic, Overview now opens in about half the time, and Agent traffic is quicker too. The files agents ask for, the pages they request most and the agents new this period are worked out in fewer passes over your data. The numbers are the same as before.
-
A reduced-motion option
Under Appearance in Account settings, choose Reduced motion to stop the app's movement: panels appear at once, charts don't animate, and live pulses and the agent ticker hold still. Match device follows your computer or phone's own “reduce motion” setting, and the website follows it too.
-
Rate Ghostie's answers
If your organization is a design partner, each answer Ghostie gives from your data now has a thumbs up and down. Tell us what was wrong or missing, and choose whether to include your question and the answer, so we can make Ghostie better.
-
A livelier Agent activity, beside the rest of the page
Agent activity on All sites now opens with the last hour in one line: requests, agents, how many your sites blocked and how many were impostors. New requests light up as they arrive, and blocked ones are marked. On a wide screen it moves into a column on the right that stays in view while you scroll.
-
Check every release with your Ghost Agents
Deploy checks run your Ghost Agent tests and a readiness scan straight after each release, from GitHub Actions or any other CI, and fail the deploy if a journey like checkout stops working for AI agents or readiness drops below the minimum you set. Create a key that can start deploy checks under Organization settings, API and MCP, and copy the step shown there. Results show on each site's Ghost Agents page. Included in the Growth plan and above.
-
Orders per 100 AI visitors compares like with like
The Revenue page could show more than 100 orders per 100 AI visitors, because orders were credited from more AI assistants than visitor counting recognised, and from utm_source tags that never show up in your logs. Visitors from DeepSeek, Grok, You.com and assistant subdomains are now counted too, and the figure sets only orders that followed a link from an AI answer against those visitors.
-
Loading sample data again replaces it
Pressing “Load sample data” a second time used to add another copy of the sample traffic, but not of the sample orders, so traffic and revenue stopped lining up. It now replaces the sample that's there.
-
Logs and orders can be sent as formatted JSON
A single JSON object sent to your ingest or orders URL over several lines, as many tools format it, was rejected. Formatted and one-line JSON are both accepted now.
-
Competitors with short names get their share
A competitor with a two-letter name, such as H&M, always showed 0% share on AI visibility even when answers named it. Short names now count when the answer names them exactly or names their website.
-
Monthly report from All sites, and company marks beside agents
All sites now has a Monthly report button for a one-page leadership summary you can print, save as a PDF or have emailed. Agents carry a small mark for the company behind them, the next setup step says how long it takes and what it unlocks, and before traffic is connected you'll see an example of the charts to come.
-
A high-contrast option
Under Appearance in Account settings, choose High contrast for darker text, stronger lines and a bolder highlight around the selected control, in the light or the dark theme. Match device turns it on whenever your computer or phone's own “increase contrast” setting is on, and the website follows that setting too.
-
Help for whoever connects your logs
The setup link you email to a developer or agency now answers their usual questions: whether it's safe, which option to pick, how to send a test request when nothing shows up, and what to do if the private URL leaks.
-
Compare your AgentScore with competitors
Readiness now has a Competitors page. Add up to three competitor websites and see every AgentScore side by side: by category, check by check, where they beat you and where you lead. Competitors are rescanned whenever your site is, so the comparison stays current.
-
Tidier Account, Organization and Site settings
Each setting now explains itself on the left with its controls on the right, so the pages are easier to scan. On Site settings, replacing the ingest URL and deleting the site are separate, each saying what happens. The activity log groups changes by day.
-
A clearer Team page
The Team page now puts your people front and centre, with each person's role, site access and last activity on one line, and how many seats you're using. Inviting people, join links, company email domain and your plan sit alongside, and each way to join shows whether it's on. Security shows how many people have two-step sign-in turned on.
-
Readiness rescans itself, and tells you when the score drops
Your AgentScore now reruns every week on its own, or every day if you choose, from the Readiness page. If a scan scores lower than the last one, or a check that passed now fails, you get an alert by email or Slack naming what got worse, usually a recent change to the site, its bot protection or robots.txt.
-
Choose which emails you get
New accounts now get a few short emails with tips while setting up, skipping anything already done, and a reminder before a trial ends. Choose whether you get those and occasional product news under Account, Email preferences; every one also has an unsubscribe link. Account emails, alerts and reports aren't affected.
-
A fuller first look when you add a site
Adding a site now shows how much of your home page AI agents can read without JavaScript and what structured data it declares, and warns when agents may see a blank page. You can also load a sample week to explore straight away, and paste a full link: we keep just the site address.
-
A tougher, fairer AgentScore
Scans now check up to six more pages from your links and sitemap, and your worst page pulls the score down. Product prices, stock and names must match everywhere agents read them. A serious failure, such as agents being blocked, now caps the whole score, and checks your site stops from running count against it. Your next scan may score lower without anything on your site changing, and we re-ran our store study on the new scoring: 255 stores, a median of 80 and 51% scoring 80 or more. Scan reports compare your score with it again.
-
Plans now include what they list
Journeys and the monthly PDF report are on the Team plan and above; Revenue and the REST API and MCP server on Growth and above, as the plan page has always listed. On other plans those pages show what you'd get. Nothing is deleted: orders and agent visits keep being recorded, report recipients and API keys are kept, and everything works again as soon as you upgrade.
-
Pages print cleanly
Printing any page, or saving it as a PDF, now gives just the page: no sidebar, top bar or buttons, dark text on white paper even in the dark theme, and charts and tables kept whole across page breaks.
-
Journeys follow long agent visits to the end
Agents that browse like a person load dozens of images, stylesheets and scripts with every page. Journeys used to look at only the first 30 requests of a visit, so a checkout or order reached later could go uncounted and “Where agents give up” could show the wrong page. Each visit's furthest step and last page now come from the whole visit, and the pages shown leave those files out.
-
Revenue at risk no longer overstates quiet stores
Mission Control's Revenue at risk, and the briefing and Ask Ghostie with it, averaged a store's agent-driven revenue over the days since its first order in the last 28 days, so a single recent order could look like a big daily figure. It now averages over the full 28 days, or over the days since a newly connected store's first order.
-
New or running Ghost Agent journeys no longer mark a site as needing attention
On All sites, a journey that hasn't been checked yet, or is being checked now, used to count as failing: the site showed “Needs attention” and its Journeys count turned red. Only a journey whose latest finished check failed counts against a site now.
-
AgentScore never presses “Buy now” to fill the cart
To see a checkout that only opens with something in the cart, AgentScore presses a product's add-to-cart button once. It now always picks “Add to cart” over a “Buy now” button that comes first, since “Buy now” can open an express or wallet checkout. If “Buy now” is the only buy button, the checkout field and guest checkout checks are marked not tested, and the report says why.
-
Ghost Agents check buttons inside embedded checkouts
Ghost Agents that work from screenshots now check buttons inside embedded frames, such as a checkout or payment provider's frame, the same way as buttons on your page. They won't press “Place order” or “Pay” there either, and if a payment or checkout frame can't be read, they don't act in it at all.
-
Clear sample data without losing real data
A site's Settings now has “Clear sample data”, which removes just the sample traffic and orders and leaves your real data exactly as it was. “Clear all traffic and order data” now asks before deleting everything.
-
We're now Ghost Agent Lab
Our name is now Ghost Agent Lab, without the "s". Nothing else changes: your sites, reports, alerts and sign-in all work as before. Emails now come from Ghost Agent Lab, and if you set up two-step sign-in again, your authenticator app will show the new name.
-
A dark theme
Ghost Agent Lab now comes in dark as well as light. It matches your device's setting to start with; choose Light, Dark or Match device under Appearance in Account settings, or from the account menu at the top right.
-
We watch your robots.txt and llms.txt
We now read each site's robots.txt and llms.txt every day. Access policy has a change history showing what changed and which AI agents it affects, and if an edit starts blocking ChatGPT, Claude, Perplexity or another major agent your policy allows, you get an alert the same day instead of finding out from lost traffic.
-
Connect Fastly, Akamai, CloudFront, Netlify or WordPress
Agent traffic can now come from Fastly real-time logs, Akamai DataStream 2, AWS CloudFront (through Amazon Data Firehose) and Netlify log drains, as well as Cloudflare and Vercel. Sites on WordPress without a CDN can install our plugin instead. Each one has step-by-step setup under Data sources, and the setup link you send a developer includes them too.
-
A shorter All sites page
The deeper views on All sites now sit in two sections, “Ghost Agent journeys” and “Agent traffic”, below your sites and journeys. Each starts closed with a one-line summary, like the busiest agent or how many shopping runs reached payment. Open one with a click; it stays open next time.
-
Impostors no longer count against a real agent's robots.txt record
When a scraper calls itself GPTBot or ClaudeBot and fails verification, the pages it asked for no longer make the real agent look like it ignores your robots.txt. “Keeping to robots.txt”, the Agent directory and Mission Control now judge each agent on its own requests. Impostors are still listed under Verification.
-
AI visibility checks always finish
A check whose questions finished at the same moment could stay “in progress” and never show its results. Every check now completes once its last question is answered.
-
Enterprise traffic history is kept for two years
Enterprise workspaces now keep agent traffic for the full two years their plan includes, instead of 400 days, and logs up to two years old can be backfilled.
-
Ghost Agents no longer follow redirects off your site
If a link, your checkout or a test's start address redirects to another website, the Ghost Agent now goes back to your site instead of carrying on there, and the step shows as blocked. A start address that redirects away ends the run with an error, so set it to the address customers land on.
-
Ghost Agent price checks read European prices correctly
Cart price checks now read prices like “1.299 €”, “1 234,56 €” and “CHF 1'234.50” as written. Before, a price with a dot between the thousands could be read 1,000 times too small, so a “total at most” check could pass when it shouldn't have.
-
Safer invitations, alerts and billing
People who join by invitation now confirm their email like everyone else, since the person who invited them can see the link too. “Being blocked” alerts no longer fire when your site turns away impostors pretending to be an AI agent. Deleting an organization now cancels its paid plan straight away, so billing stops.
-
A REST API, and an MCP server for Claude
Owners and admins can now create read-only API keys under Organization settings, API and MCP. Use them to pull your sites' AI traffic, revenue from AI, readiness, journeys, fixes and AI visibility into your own dashboards, or connect Claude through MCP and ask questions like “which agents failed checkout this week?”
-
Heavy pages are flagged, and big MCP server descriptions count
When a product, pricing, cart or checkout page is over 5 MB, the scan reads its first 5 MB and now says so in the report, so you know why product data or a price further down might be missed. An MCP server description over 1 MB is now found instead of reported as missing.
-
A slow page no longer counts as blocking AI agents
If your home page or a product, pricing or cart page takes more than 15 seconds to answer one of our AI agent test requests, the scan now leaves that request untested instead of reporting that your bot protection blocks AI agents. If every request times out, the check is marked not tested.
-
See what changed since your last visit
All sites now tells you what happened since you last looked: new alerts, journeys that ran, rescans and new agents. It suggests questions to ask Ghostie about what's on screen, shows how readiness and the Ghost Agent pass rate are trending, and lets you start your first test from a ready-made journey like Checkout or Sign-up.
-
See where Ghost Agents drop off on the way to checkout
The all-sites home now stacks your Ghost Agent runs on shopping journeys as a funnel, from landing to ready to pay. It names the step that lost the most runs, the page they most often stopped on, and what usually causes it, with a run to replay.
-
Check any IP, and see how spoofing changes day by day
Verification can now check any IP address against the agent it claims to be, the same way your logs are checked. It also shows how each request was checked, spoofing by day with spikes called out, why some requests can't be verified, and lets you export the evidence for your security team.
-
Ask Ghostie answers faster
Ghostie, the assistant in the top bar, now runs on its own smaller model made for conversation, so answers start sooner. If that model is ever unavailable, Ghostie switches to the larger one Ghost Agents use, so you still get an answer.
-
See which kinds of AI agent your site shuts out
Open a Ghost Agent journey and press “Compare agent styles” to run it as different kinds of AI agent: one that reads the page, one that doesn't run JavaScript and, where set up, one that sees only the screen. The journey says which kinds couldn't finish, and the all-sites home adds them up across your journeys. These runs don't affect pass rates or alerts.
-
AI visibility shows why AI leaves you out
When Claude rarely names you, AI visibility now lists the likeliest reasons, such as robots.txt blocking Claude's agents or product details that need JavaScript, each linking to the fix. Share of answers also shows its trend over recent checks.
-
Sort your sites, and see each one's health at a glance
The Sites table on All sites now gives each site a one-word health, shows when traffic last arrived and when it was last scanned, and sorts by any column. Click a number in the top row to see which sites drive it. You can rescan a site from its row, hide the setup checklist until you're ready, and Open alerts now tells you if no one would be notified.
-
A monthly executive report, as a PDF
Reports now has a monthly executive report: the last 30 days on one page with a short summary, the key numbers and a trend chart. Choose who gets it as a PDF on the 1st of each month, including people outside your team. When Claude writes the summary, every number in it is checked against the report's figures.
-
See which AI companies send your agent traffic
The all-sites home now groups agent traffic by the company behind it, like OpenAI, Google or Anthropic, sized by requests. It says how much comes from the biggest one, and how each company's share changed since the period before.
-
AgentScore now checks your checkout form
Most checkouts only open with something in the cart, so AgentScore couldn't see them. It now adds one item to the cart, the way a shopper would, and checks the checkout's form fields and whether guest checkout is offered. It never fills in the form or places an order. If it can't add an item, for example because a size must be chosen first, the report says why.
-
AgentScore says when your bot protection blocks it
If your bot protection turns AgentScore away even when it visits as an ordinary browser, the scan now stops and tells you so, with what to check, instead of scoring the block page. Readiness shows why the latest scan didn't finish.
-
More room on wide screens
Every page now uses the full width of your screen, so dashboards and tables show more at once. On very wide monitors the page stays centred so it's still easy to read.
-
Slow sitemaps no longer count as missing
If your sitemap is too slow to download in full, the scan now uses the part that arrived, instead of reporting that you have no sitemap, and still finds your product pages from it.
-
AI visibility (beta)
See how AI answers the questions your customers ask. Track up to 25 questions, like “best waterproof hiking jacket under $200”, and we ask Claude each one three times with web search on, every Monday or when you press Check now. You see whether it names you, where, who it names instead and which pages it cites, compared with up to three competitors.
-
Journey status on Mission Control
Mission Control now lists every Ghost Agent journey across your sites with a dot for each recent check, broken ones first, so you can see at a glance whether AI agents can still get through checkout and whether a failure is new.
-
Agent traffic: dig into a day, see page freshness, export
Click any day on the Agent traffic chart to see which agents came, what they read and what they got back. A new table shows when AI assistants, search crawlers and training crawlers last read your top pages, and Export CSV downloads every agent request in the period. Before your logs arrive, the page now shows an example week and a quick way to send setup steps to your developer.
-
Large sitemaps and robots.txt files are read correctly
A sitemap over 10 MB now passes when it's valid, and the scan still uses it to find your product pages; only one over the sitemap limits (50 MB or 50,000 URLs) is flagged. A robots.txt over 512 KB is no longer read as if it ended there, so Access policy won't offer a file to publish that drops your later rules.
-
See what moved your readiness scores
Mission Control now has a Score movers strip: the sites whose readiness rose or fell most since their previous scan, with the checks that started passing or failing.
-
A clearer All sites page
All sites now opens with one status line instead of several empty cards, points you to the single next setup step, shows how each readiness score moved since its last scan, and explains what traffic will unlock before it's connected. It also has the date range now, so you can look back as far as your plan allows.
-
Get alerts as push notifications
Alerts can now reach you as notifications on your phone or computer, even when Ghost Agent Lab isn't open. In Account, under Your alerts, choose “Turn on for this device”, then pick Push for the sites you want. Critical alerts arrive straight away, and a notification updates in place when an alert gets worse or clears. On iPhone, add the app to your Home Screen first.
-
Put Ghost Agent Lab on your phone's home screen
You can now install the app on your phone and open it full screen, like any other app. On Android, open the browser menu and choose Install app. On iPhone, tap Share in Safari, then Add to Home Screen. On a phone the menu now slides over the page, and if you lose your connection you'll see a clear offline page that reloads by itself once you're back online. Nothing about your sites is stored on the phone.
-
One place to switch sites
Switch sites from the site name at the top left of the page. It lists every site with its AgentScore, opens All sites, and adds a site. The sidebar no longer has its own switcher, so it has more room for your pages.
-
Impostor watch across all your sites
The all-sites home now shows how many requests faked a known agent's name, how that compares with the period before, and which names are being faked most, with how many different senders were behind them. Click a name to see the evidence and get a rule to block them.
-
AgentScore: fairer bot protection checks
If your bot protection checks where requests really come from, AgentScore no longer marks you down for turning away its test requests, which come from our servers rather than the AI companies'. Those checks are now skipped and explain why. Turning away a request falsely claiming to be Googlebot no longer counts as blocking AI agents.
-
AgentScore recognises more stores
AgentScore now finds your products from the product cards on your home page, whatever your product links look like, so more stores get the product and checkout checks they should. Prices are also recognised in more currencies, including zł, Kč and kr. The store benchmark under your score now uses our updated study of 258 stores.
-
Agent traffic: live requests, bandwidth and comparisons
Agent traffic now shows the latest agent requests as they arrive, which agents use the most bandwidth and roughly what it costs, and who asks for each of your top pages. Compare the chart with the previous period, and see at a glance which log source the data comes from and whether it has gone quiet.
-
Claude drafts your llms.txt
On the Access policy page, “Draft with Claude” reads your home page and key pages like About, Shipping and Returns, and writes your llms.txt: what you do, the facts an AI assistant should know when it recommends you, and links to your most useful pages. It only links to pages on your site and tells you what to check before you publish.
-
See how each agent behaves
The Agent directory now flags agents that ignore your robots.txt and ones that are new this period, shows how many requests each AI company makes for every visitor it sends you, and adds what your site sent back to each agent. Sort the list by requests, recency or spoofing, and export it as a CSV.
-
Notices from our team
When there's something you should know, like planned maintenance or a new feature, it now shows as a banner across the top of the app. Dismiss it with the × once you've read it.
-
Plans and a 30-day free trial
There are now five plans: Starter, Team, Growth, Scale and Enterprise. New accounts start with a 30-day free trial of Starter. “Plan and usage”, in the account menu, shows your sites, seats, agent requests and Ghost Agent runs this month against your plan, and compares every plan.
-
A quicker sign-in
The sign-in page remembers whether you last used Google or your email. Signing in, creating an account and resetting your password now let you show your password while typing and warn when Caps Lock is on. Two-step codes now send as soon as you've entered all six digits. It also shows a live feed of AI agents visiting the sites we watch, with no customer named.
-
Agent checkout rate and robots.txt rule-breakers on Mission Control
Mission Control now shows how often AI agents that reach a cart get to checkout, with the page where most stop, and how many agent requests ignored your robots.txt, with whether your access policy is live on each site.
-
See how your score compares with other stores
Scan reports now compare your AgentScore with the median of 234 online stores from our published study. Shared public reports can also estimate what the findings may cost a year, from the online revenue you enter.
-
Track fixes from Mission Control
Mission Control now shows how many fixes were confirmed in the last 30 days across all your sites, what's still open, and the next fix worth starting, with quick wins first.
-
Every new agent, with who's behind it
All sites now lists each AI agent that visited for the first time this week: the company that runs it, whether its requests were verified, how much it did and on how many sites. Agents we don't recognize are marked so you can check them before allowing them.
-
See which agents visit most
All sites now ranks the ten busiest AI agents and crawlers across your sites, with how far each one moved since the period before and which ones are new.
-
Sturdier readiness scans
Product and pricing pages on www are now checked against www's own robots.txt, so a rule there that blocks AI assistants is reported. A sitemap too large to read in full is now flagged instead of passed, and a scan of a very slow or heavy page stops after 2 minutes instead of running for ever.
-
Ghost Agents won't press “Buy now”
“Buy now” and “Buy it now” buttons can place an order straight away with a saved card, so Ghost Agents now treat them like “Place order” and never press them. To test a buying journey, have the agent add the product to the cart and go to checkout instead.
-
See how your readiness compares
Mission Control now shows where your average readiness score sits among every site AgentScore has scanned, such as Top 20%. Other companies' sites are never shown, only the comparison.
-
Extra Ghost Agent runs show on your plan
When we add extra Ghost Agent runs to your account, for a pilot or a busy month, they're now part of your monthly allowance. The Team page shows how many you have and until when.
-
Claude explains your fixes
On the Fixes page, press “Explain with Claude” under any fix. Claude reads the finding or the failing Ghost Agent run and writes what's wrong, what it costs you, the steps to fix it and, where it can, the change itself, ready to copy and hand to your developer. Explanations are drafts, so check them before you ship.
-
Clearer Overview, Traffic, Agents and Verification pages
Overview opens with a written summary and shows how many pages each AI company reads for every visitor it sends you. Agent traffic groups assistant visits into the questions people asked. The Agent directory lists every known agent by company with its robots.txt status. Verification shows why a request was flagged and gives you a ready-made block rule with a preview. Adding a site now takes a first look at it as you type. Each page's help is updated to match.
-
Spot new AI agents as they arrive
Mission Control now counts AI agents that visited any of your sites for the first time this week, and names the newest so you can see what it did.
-
What training crawlers cost you
Mission Control now estimates what serving AI training crawlers costs a month. They copy your pages to train models and send no visitors back. It's worked out from the response sizes in your logs.
-
Organization settings
Rename your organization, choose the role new people start with, hand over ownership, or delete the organization, all from one page.
-
Sold-out products no longer cost you points
If the product AgentScore checks is sold out, it now tries up to two other products, as a shopper would. If they're all sold out, the add-to-cart check is skipped instead of reported as a missing button.
-
AgentScore reads your store in 19 languages
Add-to-cart buttons, pop-up close buttons, cart, checkout and pricing links are now recognised in 19 languages, including German, French, Spanish, Japanese and Chinese. Stores outside English-speaking markets get a fairer, more accurate score.
-
Suspend someone instead of removing them
Pause a teammate's access straight away while keeping their role, site access and history. Restore them later exactly as they were.
-
Site access limits stay in place when sites are deleted
If someone is limited to some of your sites and all of those sites are deleted, they now see no sites until you give them access again on the Team page. Before, they could suddenly see every site.
-
Journeys: follow agents towards checkout
See how far AI assistants and autonomous agents get on your site, from product page to cart, checkout and order, the pages where they give up and the likely reason, with links to Ghost Agent tests that fail at the same step.
-
Weekly report by email
Each site has a Reports page that sums up the week with AI agents in plain words: traffic, visitors and revenue from AI, readiness, tests, fixes and the next step. Subscribe to get it every Monday, or print it as a PDF.
-
Access policy: choose which AI agents get in
Allow, limit or block AI agents by purpose (search, assistants, agents, training), with exceptions for individual agents. See the traffic, visitors and revenue at stake before you change anything, then copy a ready-made robots.txt and a drafted llms.txt.
-
Fixes: a ranked to-do list that re-tests itself
Each site now has a Fixes queue: every readiness finding and failing Ghost Agent journey, ranked by what it's worth for the effort, with quick wins marked. Assign an owner, mark it shipped, and we re-test it straight away to confirm it worked.
-
Revenue: what AI is worth to your store
Each site now has a Revenue page showing how much revenue came from people sent by AI answers and from orders AI agents placed, its share of all revenue, the average AI order, and the evidence behind every credited order.
-
More on Agent traffic: files, robots.txt and timing
Agent traffic now shows which files agents ask for (robots.txt, sitemaps, llms.txt and more) and flags the ones they can't find, lists agents that ignore your robots.txt, and has its own day-and-hour heatmap. Switch the main chart to “Share of day” to see how the mix of visitors is changing.
-
Ghost Agent tests can check the cart
A test can now pass only if the cart ends up with enough items, the right size or colour, and a total under your limit, read from the store's cart rather than the agent's word. A run that submits a payment form always fails.
-
See when agents visit
All sites now has a day-and-hour heatmap of agent requests, shown in your own time zone. Filter it to AI assistants, search crawlers or training crawlers to see each one's rhythm, and plan releases for quiet hours.
-
A live agent feed for the big screen
Open Live feed from Mission Control for a full-screen view of every AI agent request as it arrives, with impostors and blocked requests flagged. Made for a TV or second screen.
-
See what broken agent journeys cost
Mission Control's new Revenue at risk tile prices each failing Ghost Agent journey in agent revenue lost per day, and the briefing puts the costliest one first.
-
Agent-driven revenue
Send your store's orders (Shopify's Order creation webhook works) to see how much revenue AI answers and AI agents bring in. It shows on Mission Control. Find your orders URL under a site's data sources.
-
More ways for your team to join
Share a join link that lets people ask to join, or let colleagues with your company email address join or ask. You approve each request and choose their role.
-
Invite several people at once
Paste up to 20 email addresses into the invite box, separated by commas or new lines, and get a link for each person.
-
A checklist for new teammates
People who join your organization get a short Get started list on the home page: add their name, confirm their email, turn on two-step sign-in and choose alert emails.
-
Two-step sign-in
Protect your account with a code from an authenticator app, with recovery codes in case you lose your phone. Owners can require it for everyone.
-
Admin role and per-site access
Admins can manage the team without being owners. You can also limit members and viewers to the sites they work on.
-
Your plan and usage
The Team page shows your plan and how much of this month's sites, log requests and Ghost Agent runs you've used.
-
Search all help, and tell us what's missing
Press Ctrl K (⌘ K on a Mac) on any page to search every page's help, the glossary and every readiness check. At the bottom of each page's help, tell us whether it helped, and what was missing if it didn't.
-
Help on every page
Click ? beside any page title, or press the ? key, for help with what you're looking at and what to do next. Small ? icons explain buttons and numbers, readiness findings say why they matter and who usually fixes them, and there's a guided tour.
-
Mission Control
A live view of all your sites on one screen, with an agent radar, a briefing of the top actions and a readiness leaderboard. Made for a second monitor or a team TV. Click the logo to open it.
-
Viewer role
Invite people who only need to see results, like executives or an agency. Viewers can see every site, report and alert, but can't change anything.
-
Activity log for owners
See who changed what: invitations, role changes, and changes to sites, alerts and Ghost Agent tests. Export it as CSV.
-
Choose your own alert emails
Pick which alerts from each site are emailed to you, separately from your team's shared channels.
-
See where you're signed in
Your Account page lists your active sessions, and you can sign out of any of them.
-
AgentScore checks your cart and checkout
Readiness scans now check whether agents can reach your cart and checkout, whether fields are labelled, whether guest checkout is offered, and whether a CAPTCHA blocks them.
-
Our agents sign their requests
The readiness scanner and Ghost Agents now sign every request with Web Bot Auth, so your bot protection can verify it's really us.
-
Team invitations
Invite teammates as owners or members, and switch between organizations.
-
Ghost Agents
Real AI agents run your key journeys on a schedule, with pass rates and a step-by-step replay of every run.
-
Alerts by email and Slack
Get alerted when agents are blocked or a Ghost Agents test fails.
-
Agent registry
1,566 known agents, verified against their operators' published IP ranges.