Skip to main content

Goal journeys

A goal journey is one careful first-time visitor. It starts on the page you give it and clicks toward your goal, keeping a screenshot of every screen, so you see the path and where it breaks.

How it decides where to click

  1. 1

    It reads the screen

    On each screen, an AI model reads the page’s text and the links and buttons a screen reader would announce, plus any text the page made clickable without making it a link.

  2. 2

    It asks whether the goal is met

    If this screen answers the goal, the journey ends here and marks the passage that answers it. When a screen looks like the right place but nothing on it answers the goal, it first checks the link it rated most likely, and comes back to end on this screen if that page does no better.

  3. 3

    Otherwise it clicks

    It picks the link or button most likely to lead closer, clicks it, and keeps a screenshot with that click outlined.

  4. 4

    It looks inside menus

    On a desktop, the links in the site’s closed menus (Products, Resources, and the like) are options too, as they are to a visitor who points at the menu. When it uses one, it opens the menu first, so the screenshot shows where the link was. On a tablet or phone, opening the menu is a step of its own.

  5. 5

    It can search the site

    When the site has its own search box, searching can be the next step: it types a few words from the goal, waits for the results to show, and chooses among them. If nothing comes up, it types the words again key by key, as some search boxes need. It searches at most 2 times a run, and the site’s search box is the only place it ever types.

  6. 6

    A dead end isn’t the end

    When nothing on a screen looks like the way on, it opens the site map if the page links one, goes back to an earlier page and tries the next most likely link there, or takes its best guess on this one. A click that does nothing opens its link directly. It stops only when nothing left on any page it has seen looks worth a try.

  7. 7

    It clears popups first

    When a page loads, it declines the cookie banner (never accepting it) and closes newsletter, promotion, and other popups, so the screenshots show the page. One that won’t close is set aside. A popup that appears later is something it can close itself.

  8. 8

    It stays on the site

    A click that lands on another site is undone. The one exception is the site’s own link sending you to the same company under another ending (example.com to trust.example.ai), which it follows as part of the site. A link that would open a new tab opens in the same one. A CAPTCHA or bot check stops the run instead of being worked around, and a pop-up asking to confirm something is always cancelled.

Start a goal journey

On Journeys, choose New journey, keep the mode on Goal, and give it a start URL. The goal appears next, and once there’s a goal, the devices and the other settings.

The New journey dialog in Goal mode, with Desktop and Mobile picked, a start URL, a goal, and Max steps at 10.
Devices
Desktop (1280 × 800), Tablet (an iPad), Mobile (an iPhone 16), or any mix. Each device finds its own way.
Start URL
Where a visitor would arrive, usually the home page. Type it as github.com or https://github.com; https:// is added for you. Paste several URLs at once to make a row for each. To start signed in to your own site, see Teams and plans (Team and up).
Goal
What the visitor came to do, in plain English. While it’s empty, the buttons under it fill in a common one, like Pricing or Cancel a plan. See “Write a goal it can reach” below.
Max steps
How many screens it may go through, 10 unless you change it, at most 15.
Record a video
On unless you turn it off. Each run records the desktop browser finding its way, with each click shown; it adds a few seconds per screen. It needs Desktop among the devices.
Video pace
How the video plays the waiting: Real time, Brisk, Crisp. Real time keeps every pause as it happened; Brisk shortens the pause on each page to about a second; Crisp cuts it to a beat. Brisk unless you pick another. Fork keeps it.
Browse in dark mode
Asks the site for its dark theme, the way a visitor’s phone or computer set to dark mode does. Sites without one look the same. Comes with the Pro plan.
Add another journey
Start up to 5 at once, each with its own start page and goal. They share the devices and the step limit.

Write a goal it can reach

Write the goal the way a visitor would say it to a friend: one destination, described by what’s on it.

  • “Find the page that shows the Team plan’s price”

    “Pricing” Say what a visitor wants to learn or do, so it knows when it has arrived.

  • “Find how to cancel a subscription”

    “Cancel my subscription” It finds pages: it never sends a form, buys anything, or changes an account.

  • “Find whether the API has rate limits”

    “Find the docs page /api/limits” Describe the content, not the URL or the link to click. Clicking what you name tests nothing.

  • “Find the support email address”

    “Find support, then the FAQ, then contact us” One destination per journey. Start another journey for each task.

Read the result

The journey’s page fills in while it runs. Each screen is numbered in the order it was reached.

A finished goal journey: Found, scored 98 out of 100, run on desktop, tablet, and mobile. Under its video, the score for each device and Watch it as a reel, then the desktop path of two screens: the first outlines the Pricing link it clicked; the second outlines the Team plan's price in green.
The orange outline is the link it clicked. The green outline on the last screen is the passage that answers the goal, quoted underneath.
Status
How the run ended, beside the start page, the devices, and when it ran.
Runs
One chip per run, newest first. Run again adds a run and keeps the old ones, so you can compare. It waits while a run is still going.
Searches, menus, and dead ends
A screen it searched from says “Searched the site for” and the words it typed. One where it used a link inside a menu says “Opened the menu and clicked” with the menu and the link, and shows the menu open. One it gave up on says “Dead end, so went back to” and the page it returned to, which is the next screen.
Closest to the goal
On a Stuck or Out of steps run, the screen that came nearest to answering the goal is marked, with how likely it was rated to be the one. Often it’s the page a visitor should have found.
Watch it as a reel
The finished run as a short video: each screen in turn, with the camera on what was clicked and how sure the AI model was, then the answer and the score. See Organize and share for more.
Side by side
On several devices, each device’s path in a row, with how it compares to the desktop: “1 more step than desktop.” The device tabs show one device at a time.

The findability score

Every finished goal run gets a score from 1 to 100: how easily a visitor could reach the goal, read from the path the run took. A goal found in one obvious click scores 100. It’s near the top of the run’s page (under the video, when there is one) with what earned or cost points, and in the Journeys list, where the Score column sorts your goals from easiest to hardest to find.

A findability score of 63, Took some looking, with what cost points: 12 for 4 steps, 11 because the way forward wasn’t obvious at any step, 8 for a dead end, and 6 for settling on a page rated 62% likely to answer the goal.
A goal found the hard way: each line says what cost points, and how many.

A goal that was found starts at 100 and loses points for:

Steps
2 steps are free: the start page and the page that answers the goal. Each one beyond those costs 6 points, up to 36.
Dead ends
Each time it had to go back and try another way costs 8 points, up to 24. The screens the detour took count as steps too.
A way forward that wasn’t obvious
Each click or search the AI model rated less than 80% likely to lead closer costs up to 12 points, more the less likely it was, up to 30 in all. A guess, made when nothing looked right, costs the most. Those labels are where a visitor would hesitate or go the wrong way.
An unsure ending
Settling on a page rated less than 80% likely to answer the goal costs up to 10 points.
Needing the search
Each search costs 8 points, up to 16: the site’s links alone didn’t lead there. A search whose results it didn’t use, because the next click was a link the site already showed, costs nothing: not its points, not its rating, and not the screen it added. The score notes it instead.
Tucked inside a menu
When the way there was a link inside a closed menu, the score says so and names the menu. It costs nothing on its own, but the steps it took still count.

A goal that was found never scores below 40. One that wasn’t (Stuck or Out of steps) scores 1 to 30, more the closer its closest screen came. A run that failed has no score: it says nothing about the site. On several devices, each device is scored on its own and the run’s score is their average, so a phone that couldn’t find what the desktop did pulls it down.

90 and up
Easy to find.
75 to 89
Fairly easy to find.
60 to 74
Took some looking.
40 to 59
Hard to find.
1 to 39
Very hard to find, or Not found when no device reached the goal.

Scorecards: grade a site’s essentials

A scorecard asks the questions every first-time visitor has (what does it cost, how do I reach a person, where is the privacy policy) and runs each one as a goal from the same start page. In the sidebar, choose Scorecards, then New scorecard: give it a Start URL and pick the Kind of site, which decides what it looks for.

A scorecard of northwind.example: 7 of 10 essentials easy to find, 1 buried, 1 not found, and 1 that couldn’t be checked, with Pricing easier to find than last time.
Each essential is graded, with why it matters and where the run ended.
Software
What the company does, Pricing, A way to reach a person, Company name and address, Privacy policy, Terms of service, Help or documentation, Sign up, How to cancel, Security and compliance
Online store
What the company does, A way to reach a person, Company name and address, Privacy policy, Terms of service, Help or documentation, Sign up, Returns and refunds, Shipping, Accessibility statement
Business services
What the company does, Pricing, A way to reach a person, Company name and address, Privacy policy, Terms of service, Help or documentation, Security and compliance, Customer stories, Accessibility statement
Local business
What the company does, Pricing, A way to reach a person, Company name and address, Privacy policy, Opening hours, Booking

Each essential is a goal run of its own, so a scorecard uses 7 to 10 of your plan’s monthly runs, depending on the kind of site, and counts once toward the hourly limit. They run a few at a time, so it takes a few minutes; the page fills in as they finish. Each essential gets a grade:

Easy to find
Found, and the way there was clear. Found with a findability score of 75 or more.
Buried
Found, but only after detours, guesses, or a search. Found, scoring under 75.
Not found
A first-time visitor couldn’t reach it from the start page.
Not on this site
Not found, and the run read every page worth trying without any coming close, so the site most likely doesn’t have one. Counted as not found.
Doesn’t apply
A self-serve step on a site that signs customers up through sales instead. Not counted.
Couldn’t check
The run couldn’t get through (the site blocked the browser or wouldn’t load), so this says nothing about the site.

A site that sells through its sales team is graded on what it offers instead: a plans page with a way to ask for a price counts as pricing, and a demo request counts as signing up, each with a line saying so. When the only way in is a demo request, a self-serve step such as cancelling online that wasn’t found reads Doesn’t apply and isn’t counted against the site. One found anyway keeps its grade.

See the path opens an essential’s run, with every screen and its findability score. Choose Check again after you change the site: each essential that got easier or harder to find since the check before says so, and the Scorecards list counts them. Each essential has a Check again of its own too, for one that couldn’t be checked or that you just fixed; it uses one run. Essentials don’t show in the Journeys list; their scorecard lists them. Share makes a link anyone can open to see the grades, without the runs (see Organize and share).

Watch the run

A run with Record a video on has a video at the top of its page: the desktop browser moving through the site the way it did, pages loading and menus opening included. Unless its pace is Real time, the pauses while it decides where to go next are shortened, and the line under the heading says so. A cursor travels to each link or button and says what it’s about to do, like “Open ‘Pricing’”, and on a Found run it points at the answer. A strip along the bottom says Automated run, so whoever watches it, from a share link or a downloaded copy, knows an AI model drove the browser.

Watch the run: the video of the desktop browser on the Northwind home page, with a yellow cursor labelled Open “Pricing”, and two chapters beside it, the home page and the pricing page.
Select a chapter to jump to that screen. Download video saves it as an MP4.
Chapters
One per screen, with what was clicked there. The one playing is highlighted.
While it runs
Its place holds the latest screen and the screens so far. Once the last screen is in, it says it’s putting the video together, and the video takes the same spot when the run finishes. If it couldn’t be made, the run still has every screen.
Share links
Anyone with the run’s share link can watch the video too.
The reel
A different, shorter video every finished run has, recorded or not: drawn from its screenshots, with each screen held long enough to read.

Look closer

Select any screenshot to open it full size. Use the left and right arrow keys to step through the path, and up and down to scroll a tall screen.

The screenshot viewer showing step 2, the pricing page, with the Team plan outlined and the passage that answers the goal quoted below.

Problems loading a page

Every screen is also checked for what went wrong while it loaded: the page itself was an error page (like Page not found (404)), the page loaded but says it doesn’t exist (Page says it doesn’t exist, which some sites show for a missing page without an error code), one of the site’s images, scripts, styles, or fonts didn’t load, or a script failed with an error.

Nothing to report
A run with no problems shows nothing extra.
Above the screens
One line sums up the run, like “1 of 6 pages had problems”. Open it for a line per page, listed once however many devices it failed on.
On a screen
A small warning count in the card’s corner. Open the screen to see each problem under the screenshot: the file that didn’t load, or the error’s message.
Only the site’s own
Another site’s ad or tracker that fails isn’t listed, and neither is a request the site turns down on purpose (like “am I signed in?”), so what’s listed is something you can fix.

What each result means

  • Found

    It reached a screen that answers the goal and marked the passage that does.

    What to do: Check its findability score and what cost points. More clicks than you expected is a finding too.

  • Stuck

    Nothing on the last screen, or left untried on the pages before it, looked like it led closer to the goal.

    What to do: Read the labels on the last screen: that’s where a visitor gives up. If the goal was vague, choose Reword the goal on the run’s page.

  • Out of steps

    It kept clicking but used up its steps, or its time, before arriving.

    What to do: See which labels pulled it the wrong way. If the path really is long, choose Try with 15 steps on the run’s page.

  • Failed

    The run couldn’t finish: the start page didn’t open, or a bot check stopped it. The page says which.

    What to do: Check that the start URL opens in your own browser without signing in, then Run again.