Activity report
How many shoppers talked to your assistant over a date range you choose, how many used a tool, and when conversations start, compared against the period before it.
The Activity report is one of your site's analytics views: what real shoppers are doing when they talk to your assistant, over a date range you pick. Open Reports in the sidebar and choose Activity, or go to /dashboard/sites/{siteId}/reports/activity. It's the last item in the Reports section, after Overview, Conversations, Sales, Engagement, and Quality.
The report was called Insights before this rename. Its old address, /reports/insights, redirects here.
The Activity report is one page with no tabs. Two rankings that used to sit in a Demand tab moved elsewhere:
- Conversation types are the Conversation topics tab of the Conversations page.
- Purchase objections are on the Sales report, in a summary and a full ranking.
Shares one filter bar with every Analytics surface
Every report in the Reports section (Overview, this one, Sales, Engagement, and Quality), plus the Conversations list and the Quality and Engagement reports' own full-table views (Feedback, Exposures), open with the same header row: a date-range picker and, on the five reports, a compare-to-previous-period toggle (both documented once, below) and the same filter bar. The bar itself, and which of its filters a given surface offers, is documented next; every other page in this section links back here rather than repeating it.
This report has no conversation list of its own. For a filterable, searchable table of the underlying conversations, see the Conversations page. The Overview report's Conversation types card opens that page filtered on a conversation type.
Filters
Above the charts, a filter bar narrows every figure on the report to the shoppers you want to look at. It always starts with the date range (below) and a Source filter (the only two controls shown on every Analytics surface), followed by an Add a filter button and, once anything deviates from the defaults, a Reset button that shows how many filters are currently active.
Up to six more filters are available, one click away behind Add a filter:
- Agents: narrows to one or more of the site's agents.
- Agent versions: narrows to one or more specific configuration versions of your agents, for example to see how Support V3.0 performs next to V2.1 (see Agent versions below).
- Customer feedback: narrows to conversations with a Positive feedback, Negative feedback, or No feedback rating (check any combination: checking both Positive feedback and Negative feedback, for instance, shows every conversation that got a rating either way).
- Panel language: narrows to one or more of the site's enabled chat panel languages, or to Unknown for a conversation with none recorded.
- Subject or keyword: a free-text search. Add one or more terms and a conversation must match every one of them (in its subject) to stay in. This is the closest equivalent to the old search box, but it's now an added filter with its own trigger and term chips rather than an always-visible box.
- Device: narrows to one or more of Desktop, Mobile, Tablet and Unknown. See The Device filter for what each value means.
Adding one opens its panel right away; clicking a filter's own × removes it from the bar (its value resets when you do). Source narrows every figure to any combination of Web, AI Storefront, and Playground (at least one stays checked) and defaults to every conversation source except Playground, so your own testing doesn't inflate a report meant to describe real shopper activity. If you've only tested in the Playground so far, this report shows its empty state even though you have Playground history. Check Playground back on to see it.
Not every surface offers the same filters
The Quality report deliberately leaves out Customer feedback: narrowing by it would force the report's own satisfaction ratio to a constant, which defeats the point of the report. Its own full-table view, the Feedback list, doesn't have that problem and offers all six. The Sales report counts add-to-carts on your own pages, which carry none of those fields, so its bar carries only the date range and a Currency filter. The Carts page also offers Device, which reads the device each add-to-cart recorded. The Engagement report and its Starter performance and Exposures tabs read a separate event ledger with no source, agent, agent version, feedback, or language column at all, so their bar offers Device and nothing else, the one filter that ledger can answer. The Resolution, Issues, Type and Objection filters, which read the conversation labels, are offered on the Conversations list only. The Conversation topics tab and the Purchase Objection tab don't offer them either. The Topic view offers the same filters as Quality (Source, Agents, Agent versions, Panel language, Subject or keyword and Device), and the Purchase Objection tab offers only the period. A Customer feedback selection you made elsewhere is kept, not applied there, and still narrows Activity when you come back. If a filter you added on one surface doesn't apply on a surface you navigate to next (Engagement, say), the bar says so ("N filters don't apply here") with a link to see which ones and an option to drop them, rather than silently ignoring them.
Adding a filter, removing it, and which values you've picked all follow you between every Analytics surface: pick Agents and narrow to two agents here, and the Conversations list opens with the same two agents selected. This is remembered for your current browser session only (a cookie with no expiry). Closing your browser clears it, and your next visit starts back at the defaults (Source only, every source but Playground). The date range follows a slightly different rule; see Choosing a date range below.
Agent versions
Every save of an agent's configuration is kept as a saved state of a version, and every conversation stays pinned to the saved state that produced it. The Agent versions filter lets you look at one version's conversations instead of the agent as a whole, which is how you check whether a new version did better or worse than the one it replaced.
- Options. The menu lists every saved state of every version of every agent on the site, whatever its status, so a replaced version still shows up. Each option reads
Agent name · V3.1: version 3, its second saved state (see version labels). Versions are grouped by agent name, newest first within an agent. An agent you deleted no longer appears in the menu. - It combines with Agents. When both are set, a conversation has to match both: it belongs to one of the selected agents and to one of the selected versions. Selecting a version of agent A while Agents is set to agent B returns nothing.
- With no version selected, nothing changes. The filter is off by default and only narrows once you pick at least one version.
- Playground versions are hidden by default. Shoppers are only ever served the live version of the agent that answers on the site. Any other version can only be tried in the Playground, so its conversations have Playground as their source, which the default Source selection leaves out. To look at one, add Playground to the Source filter as well.
- Where it's offered. Overview, Activity, Quality, the Quality report's Feedback list, the Conversation topics tab, and the Conversations list. Sales and Engagement (with its Starter performance and Exposures tabs) don't offer it: neither has an agent version to narrow by: the adds on your pages carry none, and the engagement ledger has none. See the note below.
On the Activity report, Conversations by agent lists only the agents that own a selected version. An agent with a selected version but no conversations in range still shows, with zero.
A range reaching past retention shows a notice
Conversation transcripts are kept for 12 months. If the selected date range reaches further back than that, a notice appears above the charts naming the cutoff date. The report's figures only ever cover what's still retained, never a silently partial total for the part of the range that's already been purged. This applies to the reports built from conversations: Overview, Activity, Quality (and its Feedback list), and the Conversations list.
Last year is always partial on those surfaces, and the notice shows practically every time you pick it. The 12-month cutoff moves forward one day at a time, so in early January almost all of last year is still there, by July about half of it is, and in late December only the last few days remain. The same selection returns smaller totals as the year goes on: that is older conversations being purged, not a bug.
The Engagement report and a widget's Performance tab mostly read event data, which is kept for 24 months, so the 12-month cutoff doesn't cut those figures short and they show no notice. Two exceptions. On Engagement, Conversations started shows a dash for any range that ends before that measurement began, which today includes Last year (2025). The Engagement rate has no such gap: it reads views and opens, and shows a dash only for a range with no widget view. On the Performance tab, Messages per conversation is counted from the transcripts, so it follows the 12-month cutoff, with no notice.
Overview's engagement rate can go blank under a filter
On the Overview report specifically, the Engagement rate tile reads the same event ledger as the Engagement report: no source, agent, feedback, or language column. If the filter bar's Source selection excludes Web (the only source that ledger writes to), or if any of the other five filters (Agents and Agent versions included) narrows the scope at all, that one tile shows "Doesn't reflect the current filters" instead of a number, distinct from its usual "No widget traffic yet" empty state. The rest of the Overview report's tiles keep reflecting the filters normally. Device doesn't blank it: the event ledger records a device on every row, so the tile reports the rate over the devices you picked.
The Device filter
A conversation has no device of its own. It takes the one recorded when the shopper sent their first message, read from their browser. The reports that count conversations (Overview, Activity, Quality and its Feedback list, the Conversation topics tab, and the Conversations list) filter on that. The Engagement report and its Starter performance and Exposures tabs count events instead, and each event carries its own device.
Unknown covers more than a browser that couldn't be identified. On the conversation-based surfaces it also includes every conversation that has no recorded start:
- every AI Storefront and Playground conversation, since only Web conversations record a device;
- Web conversations from before the device was recorded, and any whose start wasn't recorded.
So a site with a lot of AI Storefront traffic shows a large Unknown bucket. That is the rule working, not a tracking fault. On the Engagement report Unknown means only that an event recorded no identifiable device, because the AI Storefront and the Playground write no events there.
The four values partition the conversations: Desktop, Mobile, Tablet and Unknown always add up to the total. Checking all four therefore restricts nothing and shows the same figures as checking none.
iPads count as Tablet
Safari on iPadOS presents itself as a desktop Mac browser. The assistant now runs a check in the shopper's page to tell those iPads from a Mac, and counts them as Tablet. Data recorded before that check shipped (see the changelog) is not rewritten: iPad visits and conversations from earlier stay under Desktop. A period that spans that change therefore shows Tablet rising and Desktop falling with no change in your traffic.
Device isn't part of the default bar. Add it from Add a filter like the others. It follows you between the surfaces that offer it for your browser session, as described above.
What it shows
Everything is scoped to the current site, the selected date range, and the filters above.
- Conversations: the number of conversations that started in the selected period. With Compare to previous period on, it shows the change against the period before.
- Assisted rate: the share of conversations where the assistant used at least one tool (a built-in chat tool like the product carousel, or a connector tool such as a Shopify catalog search) rather than answering from the conversation alone. It says nothing about orders. For orders, see the Sales report.
- Tools used by the agent: a bar chart of each tool's call count in range, so you can see which of the agent's tools actually get used. The card has two tabs. Shopper UI elements lists the tools that show something in the chat: the product carousel, product comparison, product detail and clarifying questions. Get info lists the tools the agent uses to look things up: product search, site page content, site page search, web search and web page reading. Each tool has its own bar, and each bar shows its exact count next to it, not only on hover. Each tool carries the name it has on the Tools page. Quick replies are not counted here. A connector's catalog search shows as Product search, whichever connector provides it. Other connector tools show their own name, under Get info. If nothing invoked a tool in the range, this shows an explicit empty state instead of a blank chart.
- Impressions & clicks: two bars: total recommendation impressions and total clicks on in-chat product cards, over the selected range (the same events listed in the raw Event check table under Site settings → Installation). Views of the cart card and taps on its Go to cart are not counted. Shows an explicit empty state when the range has neither.
- Peak activity: a strip with one value per hour of the day (0–23), showing when conversations tend to start. Each conversation counts in the hour it started in your time zone, and the card's description names that zone (for example "When conversations start, by hour of day (Europe/Paris)"). Hover a cell to see its hour and count. The period itself is still whole UTC days, like the rest of the report.
If the site has zero conversations at all in the selected range and source selection, the whole report shows a single empty-state message instead of a page of blank/zeroed charts.
Where the other figures moved
These figures used to sit on this report. They are on the page that owns them now, with the same filters and date range.
- Conversations over time, Conversations by agent, Tool usage and Chat panel language are on the Overview report. A period of two calendar days or fewer shows the Overview's volume chart per hour (see Charts per hour on a short period).
- Conversion rate is on the Sales report, in its revenue part. Add-to-cart rate is on its Carts page.
- Avg. messages / conversation and Avg. conversation length are on the Key figures tab of the Conversations list.
Conversion and add-to-cart rates
These two tiles moved to the Sales report. The Conversion rate tile is in its revenue part, and the Add-to-cart rate is on the Carts page of the same report. Both definitions, their time window and the states that read Not measured yet are on that page.
Choosing a date range
Click the date button above the charts to open the range picker: a list of nine named presets (Today, Yesterday, Last 7 days, Last 30 days, This month, Last month, Year to date, Last year, Last 12 months) next to a calendar for a custom range you pick by hand. Last year is the previous calendar year, 1 January to 31 December, both days included. Picking a preset applies it immediately; for a custom range, pick a start and end date directly on the calendar and click Apply. This is the same preset list used by every other date-range picker in the dashboard (Admin activity, a widget's own Performance tab).
Whenever the current range matches one of the nine presets exactly, the date button itself states the preset's name (e.g. "Last 30 days") instead of the resolved dates. That includes a range you pick by hand on the calendar: select 1 January to 31 December of last year and the button reads "Last year". A range that doesn't line up with any preset (hand-picked, or carried in a bookmarked link) shows the explicit dates on the button, since there's no preset name to state instead.
The chosen range shows up in the page URL (?from=…&to=…), so a link to the report is shareable and reopens to the same range. Without from/to in the URL (following a plain sidebar link, say), the report falls back to whatever range you last picked on this browser, and finally to Last 30 days if nothing was ever picked. How long that fallback is remembered depends on what you picked: choosing a named preset remembers it on this browser for up to a year (so it survives closing your browser); picking a custom range on the calendar is remembered for your current browser session only, the same rule the filters follow (see Filters above).
Charts per hour on a short period
When the selected range covers two calendar days or fewer (Today, Yesterday, or a custom range of one or two days), the volume chart plots one point per hour instead of one per day. A range of three days or more stays daily. The same rule applies to the volume charts of the Overview, Sales, Engagement and Quality reports. Activity has no volume chart, so this rule does not apply to it.
- The period is still whole UTC days. A one-day range is 24 hourly points, a two-day range is 48. The range is not shifted to your time zone.
- Hours are labelled in your time zone. Each point is a UTC hour, and the axis and tooltip show it in the time zone saved in your preferences, so a conversation started at 07:15 UTC reads at 09:00 for a member in UTC+2. On a two-day range the axis adds the weekday so the second day doesn't look like a repeat of the first. If your zone is offset by a part of an hour, the labels show it (for example 12:30 for UTC+5:30).
- Hours that haven't started have no value. On a range that includes today, the hours that haven't started yet have no value on the chart, not zero. Hours that have started with no conversation are zero.
- The comparison follows the same step. With Compare to previous period on, the dashed overlay is hourly too, lined up hour by hour with the current period.
Comparing to the previous period
The Compare to previous period checkbox lives inside the date-range panel, beneath the preset list, off by default. Check it to add a comparison to every KPI tile and to the volume chart (as a dashed overlay, lined up by day-of-period rather than by calendar date, or by hour-of-period on a range of two days or fewer). "Previous period" always means the immediately preceding stretch of the same length as your selected range. For a custom or uneven range (say, 5 days), that's the 5 days right before it, not a calendar concept like "the same week last month." Each KPI's badge shows the change as a compact delta (e.g. "+12%"); it's highlighted in your brand color when the metric is trending up, and stays in the neutral style otherwise: a falling number isn't automatically bad news for every metric here. The tile's footer is a fixed one-line description of what the metric measures, not a repeat of the trend or the exact compared dates.
On Last year, the previous period is the same number of days immediately before it, not the calendar year before. For 2025 (365 days), that's 2 January to 31 December 2024, because 2024 is a leap year and has 366 days. On the reports built from conversations, that whole period is past the 12-month retention cutoff, so the previous-period figures cover little or nothing and the deltas aren't meaningful. On the Engagement report, the tiles that show a figure compare against event data kept 24 months, so the earliest part of that period is purged too for most of the year, and no notice says so. While Last year is 2025, Conversations started and Engagement rate show a dash, so they have no comparison at all.
Unchecking it drops the comparison and previous-period figures everywhere. Unlike the date range, this toggle isn't remembered between visits: it resets to off the next time you open the report unless the URL still carries ?compare=true.
On the Activity report, the Impressions & clicks and Peak activity cards don't have a previous-period comparison. Only its two KPI tiles do.
What this report doesn't cover (yet)
- No revenue or cart value. Activity counts conversations and shows no order amount, revenue, or cart value. Revenue and the conversion rate are on the Sales report.
- No export. There's no CSV, PDF, or image export of this report today.
- Day grouping uses UTC, not your time zone. ⚠ Unlike the rest of the dashboard (which formats dates in your saved time-zone preference), the day buckets in the volume chart are UTC days, and the hourly volume chart on a short period counts UTC hours, only labelling them in your time zone. If your time zone is offset from UTC, a conversation just before or after midnight can land on a different day than you'd expect from local wall-clock time. The Peak activity strip is the exception: it counts each conversation in the hour it started in your time zone. That zone is the one saved in your settings; with none saved, the dashboard uses the zone guessed from your approximate location, then UTC.
Label reference
Every label the conversation judge shows you under taxonomy v1.2, what each value means, the rules it follows, and the scores and values computed from the labels.
Sales report
Two pages on one tab row. Carts shows what was added to carts, with and without a conversation, with amounts and median cart value. Orders shows revenue attributed to the assistant next to your site's, its conversion rate and average order value.