InsightsCopy Anchor Link

Insights turn the per-document AI results you already have into a single distilled output: a chronology paragraph, a relevance summary, a PII report, an investigation narrative. Instead of reading 200 rows in a field one at a time, you point Claira at the results and tell it what you want.

Every factual claim in an Insight is followed by a green-italic citation to the source document so you can verify what it's based on, and the whole output can be downloaded as a PDF that lists every source document on the trailing pages, or as an Excel workbook with each table on its own sheet.

When to use InsightsCopy Anchor Link

Common patterns:

• Chronology rollup: run a bulk scan that extracts the date and one-line summary from each document, then ask Insights to turn those entries into a single dated timeline. It comes back as a table you can download to Excel.
• Relevance review summary: after a relevance bulk scan, ask Insights to summarise what the relevant documents collectively show about the matter.
• PII or sensitivity report: run a scan that flags personally identifiable information, then ask Insights to produce a single report describing what was found and where.
• Investigation narrative: run an extraction scan for an issue (e.g. communications about a specific event), then ask Insights to weave the extractions into a coherent narrative.
• Logs and counts: ask for a table, for example documents per custodian and privilege type, then download it to Excel to sort and filter. See Tables in an Insight.

What you need before running an InsightCopy Anchor Link

Either:

• A finished bulk scan task in this case: Insights read the field the task wrote to, for the documents it finished. A task where some documents failed still qualifies: only the documents with a completed result are included, and Claira shows that count under the picker when you select such a task, or
• A document selection in Nuix Discover where the documents have values in a field you choose.

Either way, Claira reads the values from the field as they are in Nuix Discover when you click Generate Insight. It does not matter how the values got there: a bulk task, a scan in Prompt Lab, Multi-Code, Field Migration or a reviewer typing them in. If someone corrected a value after a bulk task ran, the Insight reads the corrected value, and a document whose value was cleared is left out. Claira reads with your own Nuix Discover access, so an Insight only reads values you can see yourself.

You also need a prompt describing what you want and how it should be structured.

How to run an InsightCopy Anchor Link

1. Open Claira and switch to Insights mode in the mode menu.
2. In the New insight card at the top:
   • Optionally click Quickstart to start from a template. Each tile names a common output (a chronology, a privilege log) and says in one line what it produces. Pick one to see the prompt it reads, the prompt it will put in the box, and the two ways to get its data: Run this prompt, which opens Bulk Tasks with the prompt filled in, or Pick a field instead. The tiles read the same in every case: a template is a starting point you can use as often as you like, so none of them reports whether you have used it. Use this template fills the prompt and, when a completed bulk task ran the matching prompt, selects that run as the source. The template then shows as one chip in the card, which you can remove.
   • Under Use AI results from, choose where the values come from:
     • A bulk task: pick one of your finished bulk tasks in this case. Claira uses the AI results that task wrote per document. Each option shows the task's reference (for example SC031) and name on the first line, then the field it wrote to and how many documents the Insight would read, for example 113 of 115 documents. That count can be lower than the task's own, because an Insight reads only documents the task finished writing.
     • Selected documents: a pill beside the heading shows how many documents you have selected, and you pick which field's values to use. The picker opens on the field you are already working in on this case when that field has values across your selection, and on the first available field otherwise. Generating from a field also makes it the field you are working in, so your next scan starts there. See Bulk Scan.
   • Type your prompt under Insight prompt. Describe what you want and how to structure it. For example: "Summarise these chronology entries as a single paragraph timeline, citing each event's document."
   • Under Format, choose the shape of the answer: Automatic (Claira writes text, or a table when the answer is a list of records), Text, Table or Timeline (a dated table that opens as a timeline). A template can choose one for you: the Chronology template chooses Timeline, and says so under the choices. You can still change it.
3. Click Generate Insight (or press Cmd/Ctrl + Enter).

Claira first reads the field for your documents (the button reads Reading field values… meanwhile, with a count when there are many), then starts the Insight. Documents with no value in the field are left out. If none of them has one, Claira says so and nothing is created.

If you picked a template whose prompt has never been run in this case, Claira says so and offers to open Bulk Tasks with that prompt filled in. It does not stop you generating: the source you picked may still hold something worth reading.

Claira opens the new Insight in the full-window Insight viewer straight away, so you can watch it being written. A new row also appears in the Past insights list below, with a Generating badge while Claira is working and Completed once it is done; click a row at any time to open that Insight in the viewer. Each row also shows the Insight's reference (IN followed by its number, for example IN012), and the same reference appears in the viewer's header, so two insights with similar titles are always tellable apart. Typing a reference in the search box finds its row.

If you close the viewer and stay in Insights while the Insight is being written, a notification tells you when it is ready, for example IN012 is ready to read., with an Open button. Your first Insight, and round numbers such as your 10th or 25th, also get a short milestone note: see Celebrations.

The list has a search box and a filter icon. The filter narrows by the person who created the Insight, by status, and by when it was created; the icon carries a count of how many filters are applied, and Clear filters removes them. Dates read as 2 days ago while they are recent and as a date once they are not.

Watching an Insight being writtenCopy Anchor Link

While an Insight or a revision is generating, the Insight viewer lists Claira's steps, ticking each one when it is done and timing the current one:

1. Reading the AI results, then Read 48 AI results (your number of documents). On a revision this step reads Rereading the original results, then Reread 48 original results: a revision works from the same values the first version was generated from, so your fields are not read again and the step passes quickly.
2. Writing the Insight, then Wrote the Insight.
3. Checking references, then Checked references: every reference in the finished text is checked against the documents the Insight read, and a reference to anything else is removed before the Insight is saved. This happens on every Insight.

Once Claira starts writing, the text appears below the steps a few words at a time, under the note Draft. References are checked when the Insight is finished. Draft citations are grey and can't be hovered or clicked yet, and a reference to a document outside the Insight's sources never appears. The finished Insight then replaces the draft, and its citations turn green and link to their sources. The viewer follows the text until you scroll up.

Before the first words arrive, grey lines show where the text will appear, under the note The text appears here as Claira writes it. While Claira writes, a grey line under the draft shows that more is coming, a grey bar holds the title's place until the title is ready, and the finished Insight fades in when it replaces the draft.

The draft is not kept: once the Insight is finished, only the finished Insight is saved.

If generation is interrupted, after about 10 minutes the Insight shows Generation stopped before it finished, so nothing was saved. Please try again. An earlier revision is unchanged.

With a local model, the viewer keeps its single progress line instead of the steps.

Reading and revising an InsightCopy Anchor Link

The full-window viewer has three areas:

• Revisions tab (left edge): once an Insight has a revision, a small tab appears on the left edge of the viewer. Hover it and the list slides out over the text: every revision of this Insight, oldest at the top, with the first one named Original, and the one you are reading highlighted in green. Click another revision to switch the reading pane to it, and the list steps back out of the way. Click the tab itself to keep the list open while you compare revisions, then Hide revisions to put it away. Until an Insight has been revised there is no tab at all: there is nothing to choose between, so the Insight has the full width of the window, on a wide panel and a narrow one alike.
• Reading pane (right): the prompt that produced what you are reading, then the generated text with citations inline. The block at the top is headed Original prompt on the first revision and Revision N request on a later one; it shows three lines, and hovering it shows the whole thing.
• Revise with Agent (bottom): type what you'd like changed (e.g. "Make it shorter," or "Add a paragraph about the May correspondence") and click Generate revision. A new revision appears in the revisions list, and the tab that opens it appears with the first one; the original revision is preserved so you can compare or roll back. Once the Insight has a revision chat you can open, an Open Agent chat link beside the heading closes the viewer and opens that chat in Agent Mode. This panel was headed Revise this Insight before.

A revision works from the same field values the Insight was first generated from, so every revision refines the same evidence. To pick up values that have changed in Nuix Discover since, generate a new Insight. An Insight created before Claira kept the values it read, and for which none could be preserved, can't be revised: Claira says so, and you can generate a new Insight from the same documents.

Every factual claim carries a green-italic citation naming the document it came from. Hover one to open a preview card with that document's title and the AI response that put it in the Insight, the same response you'd otherwise have to download the PDF to read. Click one and the Show sources list opens and scrolls to that document's entry. When a citation lists several IDs, each ID can be followed on its own.

Claira only ever shows you citations to documents this Insight actually read. If the model names anything else, Claira removes it from the text before you see it and reports it to us: a reference to a document that was never in the Insight is a fault to fix, not something for you to check.

Click Show sources below the body to expand the list of documents that contributed to the Insight. It shows the same thing the PDF does: a numbered entry per document with its full title, its Document ID, and the AI response that fed it in. A document the current revision never cites is tagged Not cited in this revision; it stays in the list, because the list is a record of everything the Insight read. On a long list the first fifty entries render and a button shows the rest.

Revising from Agent ModeCopy Anchor Link

You can also revise an Insight by asking in Agent Mode. The Agent never revises on its own: it shows you a Revise with Agent card with the change it proposes, and the revision runs only when you click Generate revision on it. It costs the same as revising from the panel here.

Wherever a revision was started, Claira keeps the whole chain in one Agent Mode chat for that Insight, so the sequence stays readable in one place. The chat belongs to whoever created the Insight and is shared with your case team. It is named after the Insight's reference, for example Insight revisions · IN012, and adds the Insight's title once it has one. Each entry records what was asked and what came back, never the full text of the Insight, which stays here in the viewer.

If you approve a card in a different chat, that chat keeps a short note saying where the result will appear, with an Open Agent chat button that takes you to the Insight's chat. There is no note when you approve the card inside the Insight's own chat, or when that chat belongs to someone else and has been marked Private to me.

Provenance is shown on both sides: a revision run from a chat is marked Agent in the revisions list (as is the first revision of an Insight you created from a card in Agent Mode), and the chat entry says which revision number it is. Revisions you ran from this panel are not marked: the panel is the usual place, so only the exception is flagged. A revision that fails is still recorded.

Tables in an InsightCopy Anchor Link

Claira writes a table when you ask for one, or when the answer is a set of records that share the same details: a chronology of dated events, a log, a count per custodian or per privilege type. Everything else, such as the analysis, the context and any gaps, stays as text around the table.

To ask for one, say so in your prompt and name the columns you want, for example: "List every payment as a table with the columns Date, Payer, Payee, Amount and Sources." Dates in a table are written as 2025-11-18 (or 2025-11 and 2025 when only that much is known) so the table sorts by date, and an event with no date reads undated. A table has at most six columns, so it fits a printed page; ask for two tables if you need more.

Every row carries its citations, in the cell the fact is in or in a final Sources column. They work exactly like citations in the text: hover one for the preview card, click it to open that document's entry in Show sources.

In a wide viewer a table shows as a table. In a narrow panel each row becomes a card that lists its values under the column names, so nothing is cut off or scrolls sideways.

Switching between a table and textCopy Anchor Link

Beside Revise with Agent, one button offers the other format for the revision you are reading:

• Rewrite as text when the revision holds a table.
• Rewrite as a table when it does not.

The button fills in the request below it without sending it. Edit it if you want (to name the columns, for example), then click Generate revision. The new revision keeps every fact and every citation; only the layout changes. The earlier revision stays in the revisions list, so you can go back to it at any time. You can also type the request yourself, for example "Make this a table with one row per custodian."

Seeing an Insight as a timelineCopy Anchor Link

When the revision you are reading has a table with a Date column, Claira can draw it as a timeline. An Insight you asked for in the Timeline format opens on its timeline. On any other Insight with such a table, a line above the report reads This table has a Date column, so Claira can show it as a timeline: click View as timeline, or close the line if you do not need it. The menu at the top right of the Insight, Report or Timeline, switches between the two views at any time.

The timeline shows:

• A strip of bars across the whole period, one bar per month (one per year when the case spans more than four years). A bar's height is the number of events in that period, and an empty stretch shows as a gap. Click a bar to jump to its first event.
• The events in date order, grouped by year. Each shows its date as precisely as the report gives it: a day, or Month only or Year only when that is all the documents say. Events with no date are listed last, under Undated, and do not appear on the bars.
• What happened and who was involved. The people and organisations named on an event are chips: click one to show only the events that involve them, and click it again to show everything. The filter icon above the strip lists everyone named in the timeline, and the search box finds events by any word, name or Document ID.
• The documents behind each event. Click 3 documents (the number varies) to list them, with each document's own AI result and the date it gives. Citations work as they do in the report: hover one for the preview card, click it to open that document's entry in Show sources.

When the documents behind an event give different dates, the event carries Sources disagree on the date, and the document with the other date is flagged in the list. Claira compares the event's date with the date each document's AI result opens with, as the Chronology Summary prompt writes it (2021/06/17: …), so you can check which document is right before you rely on the date.

The timeline is drawn from the report's table, so everything on it is in the report too. To change it, revise the Insight as usual: Rewrite as text removes the table, and the Insight then shows only its report. Search and filters are not saved: they reset when you close the Insight.

Finding an Insight's source documents in NuixCopy Anchor Link

If your case has a Log Field connected, Claira tags every source document with the Insight's reference (IN followed by the Insight number, for example IN012) at the moment you create the Insight. To pull the full source set up in Nuix Discover, run a search on the Log Field for that reference (for example, Log Field is IN012). The Show sources panel displays the exact reference to search for, and it also appears in the header of the downloaded PDF.

If no Log Field is connected in the case, nothing is tagged and the search hint is not shown.

Renaming an Insight and regenerating its titleCopy Anchor Link

Claira names each Insight automatically from your prompt just after you create it. You can change the title at any time from the Insight viewer:

• Rename it: click the title at the top of the viewer, type the new name, and press Enter (or click away) to save. Press Escape to cancel.
• Ask the AI for a new title: click the small refresh icon next to the title (its tooltip reads Regenerate title with AI). Claira rereads your original prompt and proposes a fresh title. This is also the way to fill the title in if automatic naming didn't complete and the row still reads Generating title….

The updated title appears immediately in the viewer and in the Past insights list, and it's the title used on the cover of the PDF and Word downloads and at the top of the Excel workbook.

Deleting an InsightCopy Anchor Link

Click the bin icon in the viewer's header. Claira asks first, says what deleting removes (the Insight and all its revisions, permanently), and keeps the dialog open while the deletion runs so you can see it working. When it is done the viewer closes and a message confirms it. If the deletion is refused, the Insight stays where it is and the reason appears under its title.

Downloading the InsightCopy Anchor Link

Click Download Insight in the Insight viewer and choose a format for the currently selected revision:

• PDF (.pdf): the report as a fixed layout, to file, print or send as it is.
• Word (.docx): the same report as an editable Word document, to drop into your own memo, brief or chronology. See The Word version.
• Excel (.xlsx): the report as a workbook, to sort, filter and annotate its tables and its source documents. See What the Excel workbook contains.

The PDF and the Word document contain:

• A header with the title, source description, and which revision it is: Original for the first, Revision 3 for a later one. In the PDF a long title takes up to three lines rather than running off the page.
• The Insight body, with citations rendered exactly as in the viewer: green italics, naming a document the Insight read. Tables print as tables, with their header row repeated at the top of every page they run onto.
• A Source documents section on the trailing pages: a prominent heading with a short note explaining that the listed documents, and the AI response shown under each, are what Claira analyzed to generate the Insight, followed by every numbered document the Insight drew from, with its title, document ID, and the value the AI read from the chosen field. When the source was a Multi-Code task, that value prints as one line per field, with the field's name in bold, rather than as the raw data Claira stores it in. An Insight with no source documents has no such section.
• A Finding these in Discover panel at the top of that section, naming the Log Field and the reference to search for, so the search travels with the file.
• A Prompt section at the very end, on a page of its own after the source documents: the prompt the Insight was generated from, headed Original, followed by each revision you asked for up to the revision you are exporting, headed Revision 2, Revision 3 and so on, oldest first. On a first revision the section holds that one prompt and is headed Prompt; with revisions under it, Prompt and revisions. The export then says what was asked as well as what came back, without putting it between you and the Insight.
• At the very bottom of every page, a line naming Claira, the report, the case, and the date you exported it, with Page 4 of 12 opposite on the right. From the second page onward, a small running header repeats the Insight's title at the top of the page, so a page read on its own still says what it is.

When you download the PDF while the Insight shows its Timeline, the PDF draws the timeline in place of the table: the events grouped by year along a line, each with its date, what happened, the people involved and its sources, which link to their entries like any other citation. An event whose documents disagree on the date names the document and the date it gives. The Word document and the Excel workbook always keep the table.

Clicking through the PDFCopy Anchor Link

Citations in the PDF body are live links. Click one and the PDF jumps to that document's entry in the Source documents section; a citation that will move is underlined, so you can tell at a glance. A citation that isn't underlined has no entry to jump to.

Each source entry carries a Cited on p. N line, and those page numbers are links back to where the document was cited, so you can read a chronology line by line and check each claim without leaving the file. Documents that fed the Insight but were never cited stay in the list, tagged Not cited in this revision.

These links work in any PDF reader and don't require Claira or Nuix Discover, which is what makes the export reviewable by someone who was only ever sent the file. If the Insight contains characters the PDF font cannot show (Chinese or Cyrillic text, or mathematical symbols such as ≥), each one appears as ¤ in the document, the last line of the document says how many were replaced, and Claira tells you after the download, suggesting the Word version, which keeps them. Accented letters, the euro sign, œ and « » print normally.

The Word versionCopy Anchor Link

The Word document carries the same header, body, Source documents section and closing Prompt section as the PDF, with a few differences that come from Word itself:

• The Insight's headings are Word headings, so they appear in Word's navigation pane, its lists are real numbered and bulleted lists you can extend, and its tables are Word tables, with the header row repeated at the top of every page.
• Citations are links too: click one (Word may ask you to hold Ctrl, or Cmd on a Mac) to jump to the document's entry in Source documents. Instead of page numbers, each cited entry has a Go to the first citation link back to where the document is first cited, because a Word document's pages move as it is edited. Entries that were never cited are still tagged Not cited in this revision.
• Every character is kept, whatever the script, so nothing is replaced with ¤.
• The footer and the page numbers are Word's own, and they update as you edit.

The filename follows the pattern Claira-Insight-<title>-rev<N>-<YYYYMMDD>.pdf, or .docx for Word.

What the Excel workbook containsCopy Anchor Link

The workbook has one sheet per part of the report:

• Summary: the title, the case, who exported it, the revision, the source, when you exported it, the prompt that produced the revision, how many documents the Insight read and how many this revision cites, and which sheets hold tables. When a Log Field is connected, it also carries the Finding these in Discover search.
• One sheet per table, named after the heading above the table, or Table 1, Table 2 and so on when there is none. The header row stays in view as you scroll. Dates written as 2025-11-18 are real Excel dates, so the column sorts and filters by date, and numbers are numbers. A Sources cell lists the bare Document IDs, ready to paste into a Nuix Discover search.
• Report: the whole revision, one heading, paragraph or list item per row, with each table replaced by the name of its sheet.
• Source documents: every document the Insight read, with its number, Document ID, title, whether this revision cites it, and the field value the Insight was generated from (one column per field when the source was a Multi-Code task). These are the values as they stood in Nuix Discover when the Insight was generated.

An Insight with no table still downloads to Excel, with the Summary, the Report and the Source documents. A value longer than an Excel cell can hold (32,767 characters) is cut, and the Summary says so. The workbook is built in your browser, like the PDF, and its filename follows the same pattern, ending in .xlsx.

Every download is recordedCopy Anchor Link

Every download, PDF, Word or Excel, is recorded for your organization's compliance records: who downloaded which Insight and revision, in which format, and when. The record is written before the file is saved, so if it cannot be saved the download is cancelled and Claira asks you to try again. The record holds none of the Insight's text.

What the model seesCopy Anchor Link

For each source document, Claira passes the model: the document ID and the value of the chosen field (the value Claira read from Nuix Discover when you clicked Generate Insight: from the field the bulk task wrote to, or the field you chose for your selection; formatted memo fields are read as plain text). When the bulk task was a Multi-Code run, that value reaches the model as one field name: value line per field, so it reads your own field names rather than the shorthand Claira stores the answers under. The model is instructed to cite every factual claim with a [doc:DOCUMENT_ID] token, which Claira renders as the green-italic citation you see in the body.

The model is always Gemini Flash 3.5 at the highest reasoning level: Insights do not expose a model picker.

Limits and known constraintsCopy Anchor Link

• Generating an Insight, and revising one, is free: it uses no tokens, whatever the number of source documents. Renaming an Insight and regenerating its title are free too. See Plans and Billing.
• An Insight reads at most 1,000 selected documents, or the documents of a bulk task of up to 25,000 documents. Either way, the values it reads must fit in a single Insight; Claira tells you when they don't.
• Claira checks that every citation names a document the Insight actually read, and removes any that doesn't. It does not yet check that the cited document supports the claim: that audit pass is planned for a future revision. Until then, treat a citation as the model's claim about where a fact came from, and verify the ones that matter through Show sources or by searching the document ID in Nuix Discover.

RelatedCopy Anchor Link

• Bulk Scan: how to produce the per-document results that feed an Insight.
• Understanding Results: interpreting the per-document AI outputs an Insight summarises.