Web research

An AI column can read the live internet as well as the row's documents: the current CEO of each company in your list, a supplier's website, today's exchange rate, the latest news about a client. Quanty builds one search per row from that row's values, picks how deep to search, and keeps the pages behind every value. A sub-sheet can go further and research a whole list per row, one verified item per sub-row, and Quanty Connect adds premium partner data to a run.

The other AI settings of a column are on Columns and AI; the column types are on Column types.

Before you start

• Web research is part of AI columns, so it needs AI to be on for your organization: a paid plan with credits (see Billing and credits). Setting it up changes the column, so it needs edit access to the table.
• It works on every AI field (Ask the AI, Summary, Classify, Find entities, Find a field, Find a date, Structured data, Sub-sheet) and on Number, Single choice, Multiple choice, Link and Email columns filled by AI.
• Searches, page reads and research runs cost credits, on top of the AI call that writes the answer (see What it costs). Official NBP exchange rates and the Polish company registries are free.
• Values of Ghost columns are never put into a search while your organization blocks the Web search channel (see Ghost and web searches).

Choose the data source

1. Open the column's settings: Add column for a new one, or Edit column in the column's menu.
2. Pick an AI type, or set How this column is filled to AI.
3. Under Data source, choose where the AI reads:

4. Set the web options that appear under it (described below) and click Create or Save.

The (i) next to Data source sums it up: "What the AI reads to fill this column: the row's documents, live results from the internet, or both. Web searches use each row's values. Mention columns with @ in the prompt to tell each search what to look for. Every search is billed from your plan's credits."

In the grid, a column that reads the internet has a small globe in its header. For a new Ask the AI column that uses the internet, the window suggests "Web results often come back as a table" with Switch to Sub-sheet: a sub-sheet keeps every result as its own row (see Researched sub-sheets).

What a search is built from

Each row gets its own search, built from the column's prompt:

• Every @-mention is replaced by that row's value. "Current CEO of @Company" becomes "Current CEO of Baltic Foods S.A." in row 1 and "Current CEO of Nordplast Sp. z o.o." in row 2.
• When the prompt has no @-mentions, Quanty adds what identifies the row (its text, or the names of its files) and a small model turns the task and that context into a short search query.
• A blank row (no documents, no text, no values) is skipped and stays pending. On a page where no row has anything yet, the top row runs from the prompt alone, so a research task can start from an empty table.

Mention the column that names the subject, such as a company, a NIP or a city, so every row searches for its own answer rather than one shared page.

Search method

Search method decides how deep each search goes.

How Auto decides

With Auto (AI picks), each row goes down this ladder and stops at the first step that fits:

1. Links in the task. When the prompt or the row contains web addresses, Quanty reads those pages directly (up to 5) instead of searching for them. Addresses on private networks are not read.
2. Exchange rates. A currency-rate task is answered from the official NBP exchange rates, free. Its source carries the badge NBP · official.
3. A quick classification. A small model reads the column's task and sorts it: quick (a Fast search), standard (a standard search), thorough (a Thorough search) or research (Deep research).
4. Fallback. If any step fails, Quanty runs a standard search, so Auto never breaks a run.

The classification is made once per column and reused for an hour, so the rows of one column normally get the same method. A method you pin yourself skips the ladder: with Fast pinned, for example, links in a row are searched for rather than read directly.

Freshness

Freshness keeps old pages out: Any time (the default), Past week, Past month or Past year. A search then uses only pages published in that window. Use it for news, prices and anything that changes.

Only these sites

Only these sites (optional) limits the search to the sites you trust. Type their domains separated by commas, for example "gov.pl, ec.europa.eu". Up to 50 domains are kept.

Polish registries

The Polish registries (GUS/KRS/CEIDG) switch "Finds the row's NIP, REGON or KRS number and gives the AI the official company record · free". It is under Data source on every AI-filled column except a sub-sheet, whatever the data source, so it also works with Documents alone. The official record shows in the cell details as a source with the badge GUS · official, KRS · official or CEIDG · official.

Researched sub-sheets

A Sub-sheet column with Internet as its data source builds a whole list for each row from live research: "The sub-sheet fills itself from live web research: one verified item per sub-row. Deep research can take a few minutes per row and is billed from your plan's credits."

1. Add a column of the type Sub-sheet (see Column types).
2. Under Data source, choose Internet.
3. In Extraction prompt, describe the list you want, with the row's subject as an @-mention, for example: "Find every logistics company that runs a warehouse in @City. For each, give its name, address, number of warehouses and a source URL."
4. Add Sub-columns for the fields you want, or leave the list empty: the AI then proposes the columns from your prompt on the first run.
5. Save and run the column.

The research finds every item the task asks for, checks each one against its sources, merges duplicates, and writes "not found" for a field the sources do not state. If a research run fails, the cell falls back to a deep search.

• With Documents + internet, a row that has documents is filled from its documents, and a row without documents is researched on the internet.
• A researched sub-sheet always uses a research run. Pinning Deep research (slow, costs more) doubles its spending cap (see What it costs).
• Freshness and Only these sites are passed to the research as instructions, with at most 20 sites.

Quanty Connect

Some answers are not on public web pages: verified company records, web traffic estimates, financial fundamentals. Quanty Connect adds premium partner sources to a run: "Premium data sources answered alongside web search in one run: company databases, web analytics, financial data and more."

Pick the sources in the Quanty Connect list under Data source (it reads No premium sources until you pick one). It appears whenever the data source is not Documents.

• Up to 5 sources per column.
• Picking any source switches the column to a research run, because partner data is reached only there. A research run takes minutes per row.
• Each partner call adds its fee, shown in the list as "~50 cr.", on top of the research run.
• If the research run fails, the cell falls back to an ordinary web search without the partner data.

On request: Crunchbase, Definitive Healthcare, Faraday AI, Harmonic, Intellizence, Kernel and ZoomInfo. They cannot be picked in the list; contact us to enable them for your organization.

What it costs

Web work is billed in credits from your plan, on top of the AI call that writes the answer. The figures below are the estimates Quanty shows, based on list prices. The charge itself follows the measured work of each request, so it can differ a little.

A small AI cost for phrasing the query and choosing the method is added to the web charge.

• Repeats are free for 15 minutes. The same search with the same settings, run again by your organization within 15 minutes (a retry, a re-run column), is answered from memory and not billed again. Research runs always run fresh.
• Web charges stay. A search is billed the moment it returns, even if the cell fails afterwards. A research run that fails or stops at 10 minutes is billed for the work it did.

Where you see costs

• Credit usage: the (i) under the web options. On a normal column: "Each row runs one web search when processed, about 13 credits from your plan on top of the AI call; deep research from ~125 credits per row. Auto lets AI pick the best method per row: reading linked pages, a quick search, or deep research." On a sub-sheet it shows the researched sub-sheet note instead. The on-request partners follow.
• Cost of running this column: the estimate in the column window (see Columns and AI). Until a column has three finished cells, it is based on a reference cell without web charges; after that it uses what the column has actually cost.
• Settings, Billing, for organization admins, under What consumes credits?: "Live web research: about 13 credits per search, ~3 per page read, deep research from ~125 credits per run." and "Quanty Connect premium sources: ~13–75 credits per partner call, on top of the research run."
• Settings, Usage, for organization admins: the Web search card, with the number of searches and what they cost. It appears once your organization has run its first search.

See where a value came from

While a web cell runs, the grid shows Searching the web… or Deep web research… and the number of sources found so far, for example "8 sources". A finished value carries a small globe after it; hover it to see how many web sources it used.

Open the cell's details (View details in the cell's menu, see Cell details, reviews and history):

• While it runs: Web research in progress, with the stage (Planning the search…, Searching the web…, Deep research in progress…, Extracting results…), the method, the query and "Found 12 sources". A research run shows entries as it finds them: "This is not the final answer. Entries may change as the run continues."
• Web sources: every page the value is based on, with its icon, title, address, publication date and a short snippet. Click a source to open the page in a new tab. Official sources carry NBP · official or a registry badge such as GUS · official.
• Query: the search text that was sent for this row, followed by the kind of fetch: search, contents (pages read directly), research or collect (a researched sub-sheet).
• Research notes: for research runs, the run's own notes, such as caveats and the filters it applied.

When you change the value by hand, the sources stay with a note: "The value was changed by hand after these pages were read. They describe the previous AI value."

Ghost and web searches

Quanty Ghost keeps sensitive columns out of channels that leave Quanty, and Web search is one of them. While your organization blocks it, a Ghost column's @-mention is left empty in the query that goes out to the internet. The AI column still sees every value when it writes the answer; only the outgoing query is filtered.

The Polish registries and NBP rates belong to the Government registries channel, which is always blocked: a column that is or reads a Ghost column skips those checks and says they were not checked. A web search you ask for in chat uses the words the chat chooses and is not filtered.

Build a researched sub-sheet from chat

You can ask for a researched list in the chat instead of setting it up by hand: describe the dataset, for example "a sub-sheet of every cold storage warehouse in Pomerania, with name, address, capacity and a source link". The assistant adds a Sub-sheet column with Internet as its data source, and can set Freshness, Only these sites and Quanty Connect sources, give the sub-columns or let the AI propose them. See AI assistant.

Good to know

• Quanty Connect and filters. With a Quanty Connect source on a normal column, Freshness and Only these sites apply only if the run falls back to an ordinary web search.
• Deep research prices. The Credit usage note quotes "from ~125 credits per row", the starting price of a research run. Deep research on a normal column without a Quanty Connect source is a single deep search, usually about 50 credits with its pages.
• Blank rows wait. On a page where some rows have values and others are blank, the blank rows stay pending until they get a value, so they cost nothing.
• Fresh pages. Quanty asks for result pages read within the last day, so a "latest price" column does not answer from a page stored days ago.

Related pages

• Columns and AI: the AI settings, running columns and credits.
• Column types: the Sub-sheet and every other type.
• Cell details, reviews and history: the cell details where sources are listed.
• Quanty Ghost: what never leaves Quanty.
• Billing and credits: plans, credits and top-ups.