Reading your first results

The ChatGPT (app) and Gemini (app) answers counted in this screenshot are demo data: fixed fictional responses run through the product's own collection code.

Overview at a glance

Overview is what you land on after signing in: how your brand stands against the competitors you track in AI answers, for the date range and filters set in the filter bar at the top of the page (28 days unless you choose 7 or 90). Everything else in the product hangs off the sidebar in four groups: General (Overview, prompts, competitors, brand perception, local results and alerts), Sources (citations and the sites behind them), Actions (To do, optimizations, page issues, articles and site health; Actions opens on To do) and Analytics (traffic, the Explorer and reports). A screen with several views shows them as tabs under its title. Which project you are looking at is set by the project switcher at the top of the sidebar, and it stays that way as you move between screens. To jump straight to a screen, choose Search or jump to in the sidebar or press Ctrl K (Cmd K on a Mac).

Every metric on Overview, from the headline to the top sources, is an Explorer result, built from the Explorer's own definitions over the Explorer's windows, so Overview and the Explorer agree for the same dates (api/services/overview.py#build_summary, api/services/explorer/metrics.py#compute_raw). A 7, 28 or 90 day window ends yesterday, and each change is measured against the same number of days immediately before it, so an answer collected today counts from tomorrow (api/services/explorer/query.py#resolve_window, #compare_window). Loading Overview spends the project's Explorer query budget, as described in The query budget and time limit (api/services/explorer/query.py#execute_queries).

The screen reads top to bottom: a headline sentence, the five headline numbers, a trend chart beside a brand ranking, then the sources AI engines cite most and the latest answers.

The filter bar

The filter bar sits at the top of every screen except the Explorer, the Analyst, settings and notifications. It holds six chips, one value each (frontend/src/lib/filters.js#FILTER_KEYS):

• Date range: the last 7, 28 or 90 days, ending yesterday; 28 unless you choose otherwise. An answer collected today counts from tomorrow (frontend/src/lib/filters.js#readDays, api/services/explorer/query.py#resolve_window).
• Engine: one AI engine, such as ChatGPT (app) or Perplexity, with every way its answers were collected pooled, so Perplexity's search and Sonar channels count as one engine (api/services/answer_filters.py#resolve_answer_filters). The chip offers the engines that answered this project's prompts in the last year.
• Tag: one of the project's tags, or No tag for prompts that carry none of them. A prompt with two tags still counts once.
• Country: one of the countries your prompts are tracked in.
• Persona: General, for answers collected with no persona, or one persona. Archived personas are listed last, marked "Archived", because their answers are still part of your history.
• Language: As written, for prompts run untranslated, or one language your prompts run in.

Tag, country, persona and language pick answers by the prompt target they were collected for (api/services/answer_filters.py#conditions). A chip with fewer than two choices in the project is disabled and says so, for example "No tag choices in this project yet", because its one value would select every answer (api/services/answer_filters.py#available_filter_options, frontend/src/lib/components/shell/FilterBar.svelte#tipFor).

A chip is enabled only on a screen where every number honours it (frontend/src/lib/nav.js#filtersFor). Elsewhere it is disabled, with the tooltip "Not available on this screen yet"; a disabled date chip still shows a 7, 28 or 90 day choice you made, because it applies again on the next screen that takes it. Where the page filters that dimension with its own control, the disabled chip says so instead, for example "Persona (page filter)" on Suggestions, with the tooltip "This page has its own persona filter"; Fact check shows "Date range (page filter)" and "Engine (page filter)", because it has its own 7, 30 or 90 day window and its own engine filter on the review queue (frontend/src/lib/nav.js#HUBS, frontend/src/lib/filters.js#disabledChipLabel, frontend/src/lib/components/shell/FilterBar.svelte#tipFor):

Screen	Chips that apply
Overview, Local, Sentiment, Reasons, Compare, Citations, Domains, Mentions, Types, Comparison, Social, Retrieved vs cited, Earned (Discover view)	All six
Competitors and a competitor's own page	Date range and engine
A prompt's own page	Date range, engine, country, persona and language
Tracked prompts	Tag, country, persona and language, which choose the prompts listed
AI traffic	Date range and engine
Alerts, Search, Site traffic, an earned source's own page	Date range

The other Prompts tabs, Objections, Fact check, Citation gaps, the Actions and Site health screens, Dashboards and Reports take none of them.

Your selection follows you from screen to screen, including through the sidebar, the tabs and Search or jump to (frontend/src/lib/filters.js#carryFilters). Switching project keeps only the date range, because tags, countries, personas and languages belong to a project (frontend/src/lib/filters.js#projectSwitchHref). Reset clears all six (frontend/src/lib/filters.js#withoutFilters). A value in a link that is not this project's, such as a tag from another project or a deleted persona, reads as "All" and is never applied (frontend/src/lib/filters.js#activeFilters).

The headline

The page title is a sentence about where you rank on brand visibility among you and your tracked competitors: "You lead AI visibility among 4 brands", "You're #2 of 4 in AI visibility", or "You're tied for #2 of 4 in AI visibility". The line under it names the gap in points: who you are ahead of when you lead, or who leads and by how much when you trail. When you trail and both windows are measured, it adds whether you closed some of that gap or it grew over the window. With no tracked competitor measured, the title reads "You're named in X% of AI answers"; with no analysed answer in the window, it reads "No analysed answers in the last 28 days yet", with the window's own length in place of 28, or "No analysed answers match the filters in the last 28 days" while a chip other than the date range is set (frontend/src/lib/overview.js#headline). When the brand at the top shows 0.0%, every tracked brand shows 0.0%, so the title reads "Every tracked brand is at 0.0% visibility in the last 28 days" instead of a rank. Values are rounded to one decimal, so 0.0% can still hide a brand named in a very small share of answers. When your visibility rests on fewer than 30 analysed answers, the line under the title reads "Low sample: n = 12 analysed answers." in place of the gap sentence (frontend/src/lib/overview.js#headline).

Rank and the count of brands include only brands with a measured visibility. Gaps and ties use the one-decimal values on screen, so two brands that both show 22.8% are tied, and the arithmetic in the sentence can be checked against the ranking beside the chart (frontend/src/lib/overview.js#rankRows).

Beside the headline, "Last run" gives the time of the latest completed answer in the project, or of a later run where Google showed no AI answer, in UTC, whatever the window; "Collecting N answers now" appears while runs are still pending or running (api/services/overview.py#_collection_status). Before the first answer completes, the title reads "Your first answers are on the way" and each section below says what it is waiting for (frontend/src/routes/(app)/+page.svelte#waiting).

Two notices can appear under the headline, one line each. "Brand figures use N of M answers; the rest are still being analysed" means some collected answers have not had their brand mentions extracted yet, so the brand numbers rest on fewer answers than the cited rate. "Failed runs in this window" lists every engine with at least one failed run in the window, as its failed runs out of its completed runs, failed runs and, on a Google engine, runs where Google showed no AI answer, which are not failures ("Perplexity 6 of 315") (frontend/src/routes/(app)/+page.svelte#notices, api/services/overview.py#_failed_runs). "Last run", the failed-runs notice and the chart's event markers describe the whole project, so the filter bar's engine, tag, country, persona and language do not narrow them (api/services/overview.py#_collection_status, #events).

The five headline numbers

The strip's header gives the window's dates, the length of the earlier period it is compared with, and how many answers were collected across how many engines. An engine counts once however many channels it answers through, so Perplexity's search and Sonar channels are one engine. Each cell has an info icon carrying its formula (frontend/src/lib/components/overview/KpiStrip.svelte#cells, frontend/src/lib/overview.js#engineCount).

Cell	What it measures
Visibility	Share of analysed answers that name your brand family (brand visibility)
Share of voice	Your brand's mentions over the mentions of every tracked brand (share of voice)
Avg. position	Your average rank in the answers that name you; lower is better (brand position)
Positive sentiment	Share of your classified mentions that are positive, on one engine channel, named under the number
Cited	Share of collected answers where one of your pages is a source (citation rate)

Positive sentiment is read on one engine channel because sentiment is compared within one channel, never averaged across engines. Overview picks the channel with the most analysed answers in the window, and the cell says "On" that channel, named the way the trend's engine lines name it ("Perplexity (search_api)" when an engine has several channels) (api/services/overview.py#build_summary, frontend/src/lib/overview.js#seriesName).

Each change is against the earlier period, in points for the four rates. An engine counts as measured in a period only when it ran (an answer, or a Google run that showed no AI answer) on at least 80% of that period's answered days, the days on which any engine has an answer (6 of 7 when every day was answered) (api/services/collection_health.py#WINDOW_PRESENCE_MIN_PERCENT). A first period with answers on only a few days is measured against those days, so a new project gets its changes as soon as it has a prior period. When some engine that answered in either period was not measured in both (an engine added to your plan during the window, such as the Google engines, ChatGPT (app) or Gemini (app) at launch, one switched off, or one that ran on only a few of the days), the change compares only the engines measured in both, while the number itself still covers every engine, and a note under the strip's header says "Changes compare only the engines measured in both periods." The brand ranking's changes follow the same rule, so the headline's comparison with the leader does too; the trend chart does not (api/services/overview.py#_compare_shared_engines, frontend/src/lib/overview.js#engineChangeNote). For Avg. position a fall is an improvement, so a lower position shows as a change for the better (frontend/src/lib/overview.js#deltaInfo). A number with nothing behind it reads "–" with "Not measured" under it and no change. A number built from fewer than 30 observations still shows, with "Low sample (n = 12)" under it, the same small-sample floor as the rest of the product.

The trend

The trend chart draws one metric across the window, one point a day unless you switch to weekly: Visibility, Share of voice, Position or Sentiment. Split chooses what each line is.

• Brands draws you and the first five competitors of the brand ranking. With more competitors than that, a note under the chart says "Showing you and the top 5 competitors by visibility" (frontend/src/lib/components/overview/TrendCard.svelte#SHOWN_COMPETITORS, api/services/overview.py#TREND_BRANDS).
• Engines draws only you, one line per engine channel with answers in the window; with an engine chosen in the filter bar, only that engine's channels. A note lists the engines with no line: not on your plan, no answers in this window, or, for Google AI Overviews and Google AI Mode, no AI answer shown in this window (every run there showed none), and only the chosen engine when there is one (api/services/overview.py#_not_measured).
• Sentiment is available only with Brands, and on one engine channel at a time, chosen from a list above the chart; it starts on the channel the Positive sentiment cell uses. Under Engines the Sentiment button is disabled, because sentiment is not comparable across engines, and it is also disabled when no engine has an analysed answer in the window (api/services/overview.py#build_summary, frontend/src/lib/components/overview/TrendCard.svelte#sentimentOff).

The menu beside Split sets the granularity. Weekly needs the 90-day window, and a week the window covers only in part says "partial" when you hover it. The same menu offers Download CSV, the Explorer export of the query the chart ran, when the project's plan includes data exports, and Open in Explorer, which opens that query in the Explorer.

How to read the lines:

• A day with no measured value is a gap in the line, never a drop to 0%. Rate axes start at 0%; the Position axis is upside down, with #1 at the top.
• Each line ends in a label with its name and value, placed on its last point with at least 30 observations, or its last measured point when none has that many. A dot on a point with fewer than 30 observations (the dot under a line's label, a point with no neighbour to join, or the point you hover) is drawn hollow, and hovering such a point says "low sample (n = N)". A line with no measured point still gets a label, with "–" for its value (frontend/src/lib/overview.js#labelIndex).
• When a label sits on an earlier point than the chart's last day (because the newer points have fewer than 30 observations), a note under the chart says so: "Labels show the latest day with at least 30 answers: Aug 10" when every label is on that one day ("week of Aug 10" by week), otherwise "Labels show each line's latest point with at least 30 answers." The count is in the metric's own unit: answers for Visibility and Position, mentions for Share of voice and Sentiment. A label that could only use a low-sample point is already the line's newest value, so it does not bring up the note (frontend/src/lib/overview.js#labelNote).
• Fewer than five days (or weeks) with any measured value, and the chart is replaced by "Not enough history for a trend yet", or by "Not enough answers match the filters for a trend" while a chip other than the date range is set (frontend/src/lib/components/overview/TrendChart.svelte#MIN_PERIODS).

Markers along the bottom of the chart show what happened in the window: articles published, optimizations applied and alerts detected, dismissed alerts left out (api/services/overview.py#events). Alerts are tinted. Markers closer than 14 pixels merge into one, and a marker's tooltip lists up to three events and how many more there are. Choosing a marker opens the first event it lists (frontend/src/lib/overview.js#groupEvents, #mergeEventGroups).

The brand ranking

Beside the trend, Brand ranking lists you and every active tracked competitor by visibility over the window, with each brand's change in visibility, its share of voice and its average position. It is built from brand mentions in AI answers, not from which domains are cited, so it answers a different question from the domain-citation figures on Competitors (api/services/overview.py#_ranking).

Ranks follow the headline's rule: brands showing the same one-decimal visibility share a rank, and a brand with no measured visibility has no rank and shows "–". Unmeasured brands sit at the bottom. A competitor's row opens its page on Competitors; your own row, marked "(you)", is not a link (frontend/src/lib/components/overview/BrandRanking.svelte#rows).

Top sources and recent answers

Top sources lists the five domains cited by the most answers in the window, each with the share of collected answers citing it ("In 18.4% of answers"). A domain is marked "Your page" when it is your project's domain and "Acme's site" when it belongs to an active tracked competitor. All sources opens Domains (api/services/overview.py#_sources). Whether a third-party page names you is shown per page in the answer drawer below, not per domain here.

Recent answers shows the five newest of the 20 newest completed answers the page loads. It is a feed rather than part of the window, so it includes today's answers, but it does follow the filter bar's engine, tag, country, persona and language. Each row shows the engine, the prompt, and "Mentioned #N" (your position in that answer), "Not mentioned" or "Not analysed yet"; a link icon means one of your pages is a source in that answer (api/services/overview.py#recent_answers, frontend/src/lib/components/overview/RecentAnswers.svelte#SHOWN). All prompts opens the prompt list.

The answer drawer

Choosing a recent answer opens it in full in a panel on the right. The header tags the engine and, when the answer was collected for one, the country, persona and language; View prompt opens the prompt. Under the prompt, one line gives when it was answered (in UTC) and how many brands and sources it has, with the same Mentioned or Not mentioned tag and a "Cited" tag when one of your pages is a source, read from the answer itself once it has loaded (api/services/overview_answer.py#build_answer, frontend/src/lib/components/overview/AnswerDrawer.svelte#status).

The answer is shown as plain text, split into paragraphs, with each named brand highlighted once, where its first mention was found. Markdown is tidied away: headings, bold and italic marks, citation markers such as "[1]" and "[2][6]", and a link's address, which leaves its text (api/services/overview_answer.py#_markup_to_drop, #_clean). A brand whose name cannot be found in the text, or that would overlap another highlight, is left unhighlighted rather than marked in the wrong place. An answer longer than 20,000 characters is shortened, and a note says so (api/services/overview_answer.py#segment_answer, #MAX_ANSWER_CHARS).

Beside the answer:

• Brands in answer: each brand in order, with its position and its sentiment in this answer ("Not classified" when there is none). Untracked brands are muted.
• Fan-out searches: the searches the engine ran while answering, when any were captured.
• Sources: up to 20 distinct pages, best rank first, each linking out. Each page is marked Your page (on your domain), Mentions you (its saved text names your brand, matching the names in your own brand family literally), No mention (its saved text does not), or Not checked (there is no saved copy of the page that could be checked, or your own brand family is not set up yet, so there is nothing to look for) (api/services/overview_answer.py#_sources, #MAX_SOURCES, frontend/src/lib/components/overview/AnswerDrawer.svelte#STATUS).

Under the lists, a Next step block appears when one applies, chosen by fixed rules, first match wins (frontend/src/lib/overview.js#nextStepMode):

1. A cited page on a third-party site, neither yours nor an active competitor's, that does not name you, when the answer was collected inside the earned sources window of the 90 days ending yesterday. Owners and editors get Save as earned source, which saves the page as an earned-source opportunity; viewers get a link to Earned sources instead (api/services/overview_answer.py#EARNED_WINDOW_DAYS, frontend/src/routes/(app)/+page.server.js#actions). An answer from today is outside that window, so it never offers this step.
2. The answer has been analysed and does not name you: Open prompt.

Previous and Next, or the K and J keys, move through the answers the page loaded (the 20 newest); Esc closes the drawer.

Where the old Overview cards went

If you want...	Go to
Citation rate by country, persona or language	The Explorer, broken down by that dimension
Citation rate engine by engine	The Explorer, broken down by engine
AI referral sessions and conversions	AI traffic
Recent alerts	The trend's markers, and Alerts
Prompts that gained or lost ground	The weekly report

What no data means

A metric with nothing behind it yet reads as no data on every screen, never as a zero: those are different states with different causes. On Overview that is "–" and "Not measured" in a cell, and a gap in a trend line. Rather than repeat the rule here, see How to read any number here on Metrics defined for the full explanation of when a metric is ready to read as a real number.

Which number to read first

Start with visibility, the number the headline is built on. Of everything on Overview, it answers the most basic question: do AI engines name your brand at all, when someone asks the kind of thing your buyers ask. Share of voice, position, sentiment and the cited rate all matter, but each of them is a more specific question than that, and none of them means much until the basic one has an answer.

Where to go next

If you want to see...	Go to
Which prompts you are tracking, and each one's own answers	Prompts
Where your domain has actually been cited	Citations
Prompts where a competitor is cited and you are not	Citations → Gaps
How you compare against the competitors you added	Competitors
Traffic AI engines are actually sending to your site	Traffic → AI traffic
Search Console and GA4 performance, if you connected them	Traffic → Search and Site traffic
Any metric broken down and filtered your own way	Explore → Explorer

Related

• Metrics defined: every number's exact formula and denominator
• Explorer: the definitions and windows behind every Overview number
• Troubleshooting: what to check when a number reads as no data
• Your first scan: what has to run before there is anything to read