> For the complete documentation index, see [llms.txt](https://docs.uprightplatform.com/llms.txt). Markdown versions of documentation pages are available by appending `.md` to page URLs; this page is available as [Markdown](https://docs.uprightplatform.com/upright-agent/using-the-chat.md).

# Using the chat

## Starting a conversation

There are three ways to open the chat:

* **From a company, group, or portfolio page** — click **Ask AI** in the top navigation. The chat opens as a side panel and inherits the context of the page you are on. A small chip at the top of the chat shows what context the chat is anchored to (for example, "Double materiality · Storebrand"). Questions like "summarise this" or "what's driving this" resolve against that context automatically.
* **From the standalone chat view** — go to [`uprightplatform.com/chat`](https://uprightplatform.com/chat) for an open-ended conversation that is not tied to any particular page. Use this when you want to compare multiple companies or ask methodology questions.
* **From a welcome suggestion** — the empty chat view shows a few suggested prompts (for example, "Net impact of Apple", "DMA results for Nordea"). Click one to start the conversation with that question pre-filled.

## Attaching files and searching the web

The composer has two optional inputs next to the message box:

* **Add files** — attach a document (for example, a PDF) and ask the chat to read it alongside Upright's data. A common use is to hand the chat your own materiality assessment, sustainability report, or stakeholder-engagement findings and ask it to compare them against Upright's modelled results.
* **Web search** — a toggle that lets the chat pull in external context from the public web (for example, recent controversies, news, or background on a company) on top of Upright's own data. It is off by default; turn it on when you specifically want the chat to look beyond Upright's dataset. Answers drawn from the web are cited like any other source so you can see where they came from.

## What a good question looks like

The chat works best when the question is specific about which company / group / portfolio you mean and which analytical lens you want.

**Good examples:**

* "Show and summarise the net impact of Storebrand."
* "Compare the DMA results for Apple, Microsoft, and Alphabet."
* "What products are driving Health and safety as a material topic for Neste?"
* "List the constituents of this portfolio sorted by net impact."

If you do not name a company explicitly and the chat does not have a page context to work from, it will ask you which company you mean rather than guess.

### Question types that work well

These are the kinds of task the chat is built for. The phrasing is illustrative — substitute the company, fund, list, or topic you care about.

* **Single-metric lookup with the "why".** "What is this carmaker's GHG-emissions impact, and which products drive it?" — the chat returns the figure and can drill into the positive and negative contributors and the per-product reasoning behind them.
* **Peer / industry benchmarking.** "How does this shipping company's net-impact ratio compare to its peers, and where does it rank?" — the chat builds a peer set and returns a ranked comparison.
* **Screening a list on a theme.** "Screen these companies for biodiversity impact" or "which EU companies are most SDG-aligned?" — paste a short list, or ask for a ranked screen.
* **Investor due diligence.** "Give a 360° view of this company for an Article 9 portfolio screen — net impact, the main risks, and any controversies." — combine Upright's lenses with the **Web search** toggle for external context.
* **Compare your own report against Upright.** Attach your own materiality assessment with **Add files** and ask the chat to compare it against Upright's modelled double materiality — where the two converge and where they diverge.
* **Site-level physical climate risk.** "Assess the physical climate risks across this company's sites." — the chat identifies the company's locations and screens each for heat, wildfire, flooding, and other hazards.
* **Working with your own results.** From your assessment's home page: "Summarise my results", "What's missing in my assessment?", or "Explain the matrix to my CFO."

## Prompting tips

* **Name the entity precisely**, or open the chat from that entity's page so it inherits the context automatically.
* **Pick one lens and ask one thing at a time.** Use follow-up questions to drill in rather than packing several asks into one message.
* **Ask "show your sources"** (or click the citation chips) when you want to see exactly what an answer is based on.
* **Paste a short list** when you want to compare or screen several companies side by side.
* **Turn on Web search** when you want external context (news, controversies); **attach a file** when you want the chat to reason over your own document.

## What to expect — where the chat is strongest

Treat this as "how to get the best results", not a disclaimer. The chat is most reliable on **entities Upright covers** and on Upright's own lenses:

* **Strongest:** double materiality, net impact, financial effects, methodology questions, and peer comparisons or drill-downs for covered companies, funds, and products — anywhere the answer comes straight from Upright's structured data with citations.
* **More variable:** companies with very thin public data, very large screens, hyper-specific single-number lookups, unusual or long product value chains, and site / climate questions where a company's locations are hard to find on the public web.

When the chat is on weaker ground it will tell you what it could and could not establish rather than inventing a number — and naming the entity precisely, narrowing the lens, or splitting the question into follow-ups usually gets a better answer.

## What a response contains

Responses are designed as analyst-style briefs — the chat does not narrate every number in the data; it summarises the headline story and surfaces the supporting numbers in structured elements next to the prose:

* **KPI tiles** at the top — typically 1–4 headline numbers (for example, NET IMPACT, MATERIAL TOPICS, REVENUE) with a coloured tone indicator. The tone is based on the underlying valence of the metric, not on whether the news is good or bad for the user.
* **Tables** below the prose for row-level data (dimensions, drivers, peer comparisons). Every table has a **Copy as TSV** and **Download as Excel** button in its title row.
* **Inline citations** — small chips next to a claim that link back to the underlying tool result.
* **Callout boxes** for verdicts, exec summaries, and "what this means" footers.
* **Deep links** — the first time the chat mentions a company or a material topic, the name is linked to the corresponding platform page (DMA dashboard, single-matter detail, lens view, etc.) so you can jump straight from the chat to a deeper look.
* **App cards** — for some questions (for example, a product value-chain query or a physical-climate-risk heatmap) the chat renders a dedicated visual card below the prose. The card is the diagram; the prose narrates it.

## Getting answers out, and languages

* **Export** — every table has **Copy as TSV** and **Download as Excel** in its title row, and a full answer can be printed or saved to PDF from your browser's print dialog.
* **Languages** — you can ask in your own language; the chat replies in the language of your question.

## Follow-up questions

Conversations are stateful. You can refer back to anything the chat just said:

* "Now compare that to Microsoft."
* "Drill into the negative environment score."
* "Why does this score so negatively on biodiversity?"
* "Show the supporting articles."

If you want to start fresh, click **New chat** (in the chat header, or in the conversation-history sidebar on the standalone view). If the side panel was opened from a specific page, the page context follows you into the new conversation.

## Citations and sources

Whenever the chat makes a claim that comes from Upright data, it attaches a citation chip linking back to the tool result. Click the chip to see exactly which data point supports the claim.

When the chat lists or cites multiple companies, the answer ends with a **Sources** footer listing each one. Companies that are part of your organization's own tailored library (rather than the public Upright library) are marked with a small lilac **Private** chip after the name.

## Private vs. public results

Some organizations have access to tailored variants of company results — for example, a CSRD assessment that uses your organization's own product list or sustainability data points instead of the public defaults. When the chat answers from a tailored variant, the company name carries a **Private** chip the first time it is mentioned. Public Upright library results are unmarked (no chip needed).

## Conversation history

The chat history sidebar lists your previous conversations. Conversations are stored against your user account and are not shared with other users in your organization.

## What to do if the answer looks wrong

The chat is grounded in Upright's structured data, but it is still an AI. If a response looks wrong:

* Check the citation chips — they link to the underlying tool result and let you verify the source.
* Ask the chat to "show your sources" or "what is this based on?" — it will surface the tool calls it used.
* If you still believe the data itself is wrong, raise it with your Upright contact person; the feedback handling policy in the [appendix](/appendix/feedback-handling-policy.md) applies.