AI prospect research with a cited source on every card

What AI prospect research should give you

Good prospect research answers four questions before you write a word: what the company does, whether it fits your offer, why now, and who owns the problem.

A chatbot summary can answer them in a minute, but it does not say where a line came from, how old it is or whether the model invented it. That matters most in cold email. One wrong detail, such as a "congrats on the funding round" that is a year late, tells the reader the message was written carelessly.

When you compare any account research tool, look for four things:

• A source on every fact, so you can click through and check it.
• Dates: when the page was read, and when the fact goes stale.
• A check on sources: a fact that cites a page the AI never actually retrieved should be thrown out.
• A fair price when a run fails: nothing charged if the report does not finish.

FoxKnock's account research tool is built around those four. A report is a dossier of evidence cards, shaped for outreach: a short summary, a "why now" note, sections on the company, news, hiring, tech stack and people, and insights such as likely pains, talking points, objections and personalization hooks. Every personalization hook needs a source URL too.

The summary, likely pains and talking points are the AI's reading of those sources, so treat them as a starting point and check the cards behind them.

Evidence cards: every fact has a source and a shelf life

Each fact in a report is stored as an evidence card. The last column shows an illustration with a fictional company: nothing about Northwind is real, and the card format is the point.

Two rules keep the cards honest.

Unsourced facts are dropped. A fact survives only if its source URL matches a source the agent actually retrieved during that run (a page it opened or a search result it received). That check stops invented sources, but it cannot prove the AI read the page correctly.

Facts expire. News, funding, hiring and event facts last 90 days from retrieval, and the rest last 365 days. Expired cards are not used to write emails. When a newer report replaces old cards, the old ones are kept and marked superseded, so an email you already wrote still shows the source for each claim.

Quick, standard or deep: how the research depths work

You choose how much work FoxKnock does per company. The prices are default credit prices (they may change), and 1 credit is $0.01 at list price.

Quick: sort a long list

Quick reads the homepage and up to 3 key pages, searches recent news (up to 6 results from the last 365 days) and detects the tech stack from page HTML, headers and mail records. Claude then writes the dossier from what was read.

Standard: for accounts you plan to contact

A Claude research agent covers what the company does, fit against your ideal customer, 12 months of news, funding, launches and leadership changes, hiring, tech stack, likely pains, why-now triggers, key people and personalization hooks. It is the default depth in the REST API and in autopilot.

Deep: for the accounts you cannot afford to get wrong

Deep adds competitors and alternatives, a customer angle (named customers, case studies, review-site sentiment) and second-source checks on the strongest claims. If you pick a key contact, it also researches that person with 6 more searches and 6 more fetches. Use it for high-value accounts, or the few where the first email has to be right.

A time limit is a ceiling, not a promise of how long a report takes, and a report that runs past it fails and charges nothing. Standard and deep research run on Claude, by Anthropic, using Claude's built-in web search and fetch tools. Quick research reads pages and news through FoxKnock's search and fetch providers, then Claude writes the dossier. Either way, page content is sent to Anthropic for the AI step.

How to research a prospect before cold email

A workable order, with FoxKnock's tools where they save time:

1. Spend on fit first. Research costs credits, so run it on companies that match your ideal customer. Lookalike discovery finds companies that resemble your best customers, and accepting a result can queue standard research for it (up to 50 at once).
2. Match the depth to the stakes. Quick for triage, standard for the accounts you will email, deep for the few you cannot afford to get wrong.
3. Read "why now" first. A recent hire, funding round or launch is a better reason to write than a general compliment. Plays can start research when a buying signal appears, and the opt-in autopilot researches best-fit companies daily, up to a limit you set.
4. Use one or two facts, not ten. One cited, recent fact reads as relevant, while a pile of facts can read as surveillance. In the Northwind example, the hiring card supports one opening line, such as "I saw Northwind is hiring its first sales development rep", and nothing more. For the manual version of this checklist, see how to research a prospect by hand.
5. Let the cards write the claims. AI-written drafts must cite a card for each specific claim about the prospect, and an unsupported claim blocks the draft. Under the default approval level you read and approve every draft before it sends (see approval-first outreach).
6. Click through when it matters. If a line names a number, a person or a date, open the source and read the quote before you approve.

AI sales research compared: manual work, a chatbot and evidence cards

• Manual research gives you the most judgment, but it is slow and its consistency depends on who does it. Notes can pile up in tabs without dates, and the work is easy to skip when the queue is long.
• A chatbot prompt is fast, but it can be hard to tell where a line came from unless you ask for sources and check each link yourself. Some chatbots with web search do show links, so check how yours behaves.
• FoxKnock evidence cards add a trail to AI research: a source URL, a retrieval date and an expiry on every card, with cards that lack a retrieved source removed before you read them.

None of these removes the need for judgment. The cards put the evidence in front of you, so checking a line takes one click.

What it costs

Research is paid in credits at the default prices above, and you are charged only when a report completes. You can also set a monthly spend cap in Billing. The plans:

• Free ($0): 500 one-time credits, granted once you verify your email, enough for two standard reports at default prices.
• Starter: $49 a month with 5,000 credits.
• Growth: $149 a month with 20,000 credits.
• Agency: $399 a month with 60,000 credits.

Unused credits roll over. Credit prices can change, so check the live numbers on the pricing page.

To see how it reads on your own market, create a free account and spend the free credits on two standard reports for companies you would really email.