Every draft now starts with research. Before the AI writes to a prospect, it checks that prospect's own website for fresh, verified facts to ground the email in, instead of writing from a generic template with a name and company dropped in. You are only charged for research that actually gets used and accepted.
What changed
When a contact enters a campaign, the AI now researches that prospect's website before drafting anything. It is not paraphrasing an old, cached summary. It pulls current facts, the kind of specific details a real SDR would look up before reaching out: what the company does, recent changes on their site, language they use to describe their own product or service, the kind of detail that signals an email was actually written for this specific prospect.
That research is verified before it goes into a draft. The system checks that a fact actually came from the prospect's site rather than being invented, so what ends up in your email is something a prospect could check for themselves and recognize as accurate, not a plausible-sounding guess that happens to fit the shape of a real fact.
Billing here is tied to outcome. You pay for research that gets accepted into a draft, not for every attempt the system makes. If a research pass does not produce anything usable, whether because the site had nothing current to pull from or the facts it found could not be verified, you are not charged for it.
This research step runs before drafting begins, not alongside it or after the fact. That ordering matters: the AI is not writing a plausible-sounding email first and then trying to backfill it with supporting detail. It gathers and verifies what it can say about the prospect first, and only then starts writing, so the facts shape the email rather than being bolted on afterward.
How to use it
There is nothing to turn on. Once a contact is in a campaign that uses this, research happens automatically before the draft is written. When you review a draft, you should be able to see specific, grounded detail about the prospect's business rather than filler language, since that detail is what the research step is meant to produce, and it should be the kind of thing that makes a prospect think an actual person looked at their site before writing.
If a draft reads generically, that is worth a second look. Check whether the contact's own website actually has usable, current content for the AI to draw from. A near-empty site, a site under construction, or one with almost no text will naturally give the AI less to work with, and the draft will reflect that limitation rather than the research step failing outright.
When you are reviewing billing, look at what you were charged for research. Because charges are tied to accepted research, a contact whose research did not produce a usable fact should not show a charge for that step, and if it does, that is worth flagging.
For lists that include a mix of well-established companies and newer or smaller ones, expect research quality to vary contact by contact, and treat that as useful signal rather than a flaw. A prospect with a thin website is often a prospect where a different angle, one that does not depend on site-specific detail, will work better anyway, and noticing which contacts consistently get thinner research can help you refine how you segment a list before a campaign even launches.
Why it matters
Prospects can tell the difference between an email that mentions something true about their business and one that could have been sent to anyone. Research grounded in the prospect's actual website is what makes a cold email feel like it was actually written for that person, which is a big part of what gets a reply instead of a delete, especially in inboxes already full of generic outreach.
Verifying facts before they go into a draft also protects you from the worst version of AI outreach: an email that confidently states something wrong about a prospect's company. That kind of mistake damages trust fast, and it is hard to recover from once a prospect has seen it. Grounding drafts in verified, current research is meant to prevent that outcome entirely, not just reduce how often it happens.
Paying only for accepted research means the cost of this step tracks value delivered, not just effort spent. If research does not produce anything the draft can use, you are not paying for a dead end, which keeps the economics of outreach honest as your campaigns scale.
This also changes what "personalization" actually means for a campaign running at volume. Personalization that depends on someone manually looking up each prospect does not scale past a handful of contacts a day. Personalization grounded in automated, verified research does, which is what makes it realistic to run outreach that still reads as specific to the prospect even across a list of hundreds of contacts.
None of this requires anything from you beyond running the campaign as usual. Research happens on every contact automatically, and the only thing that changes in what you see is the quality of what shows up in the draft once that research is available.
Over a full campaign, the difference tends to show up less in any single email and more in the pattern across a list. A campaign drafting from real, verified research should read noticeably less repetitive across contacts than one working from a template, since each email is pulling from something different and specific to that prospect rather than a shared shell filled in with a name.
Availability
Prospect website research is active now for campaigns that use AI drafting, and applies automatically as contacts move through the campaign, with no extra setup required. It runs for new contacts entering a campaign now, and any contact already mid-sequence picks it up on its next scheduled touch.