The glance: is this company worth my time?
Glance takes about fifteen seconds and answers one question: is this company worth a full run?
It is on every list of companies the studio shows you (Leads, the lookalikes on a finished pitch, a Counterprospekt report's client list) and on Research prospekts. On Leads it is the card's one press, because a shelf is for triage; Prospekt, the full run, is the purchase you make from the glance. A glance goes in your Library like any other reading, and the shelf stops offering that company; Suggested never proposes work you have already done, so what is on it is always what you have not looked at yet.
What comes back is one card: a fit score out of 100, the single argument the case rests on, two or three signals for and against, and the roles worth approaching. It reads the company's own site to do it, capped at two pages and one search, and the card says how long it took against the fifteen-second ceiling so you can hold us to it.
It comes back whole, or it says so. A reading that arrives missing part of its answer is asked for a second time and then stopped, rather than shown to you as a score with nothing under it. The one exception is the second score, the buyer reading: a glance that has its score, its argument and its signals but no second reading is kept, with that reading marked not scored rather than invented.
It knows what you sell. The glance reads your brief's facts and your fit thesis before it scores, so a target is judged as a company that might buy from you, never as a company that resembles you. A note you type on the run outranks the brief: if you know a company is a customer in the making, say so, and the score follows. The card says in one line what it made of your note, and a score that goes against it is asked for a second time: either the number moves, or that line says what the studio read that outweighed you.
It is kept. Every finished glance gets a card in the Library (the score, the one thing, the desks it named) marked Glance so it is never mistaken for a pitch. Press the card and you are back on the glance, with the same presses on it: name the people, write a piece, or Prospekt. The card also carries Prospekt → on its own, because that is the one thing a glance is missing. A finished pitch carries no button, since pressing it is what opens it. It has its own link, and it tags and filters like anything else there. Once the full pitch is finished it takes the glance's place in the list rather than sitting beside it: one company, one card. Nothing is deleted: the glance keeps its own link and is still there while the run is out, and if the run fails. Run a company a second time and the newer run takes the card the same way; the older run keeps its link.
Two pages buys the front of a business, not all of it. Where a company plainly runs more than one line, the card says which one it scored: "Scored as a consumer app; it may also have a B2B channel this glance did not read." That line is the reason to buy the full run rather than to trust a low score.
It names roles, never people. Naming a person needs a source, finding a source needs searches, and searches are what a fifteen-second answer does not have. Naming them is the next press.
Outside your brief, and still a buyer
Every company gets two readings, and they answer different questions. The score at the top of a pitch is how well does this match the brief you wrote. Beside it the studio asks a second question it does not show unless the answer surprises: would this company buy what you sell, whoever your brief names.
Most of the time the two agree, and you see nothing extra. They come apart in one direction: a company your brief does not describe, which would buy the work anyway. Then the number at the top of the pitch is the buyer score, labelled a buyer, outside what you hunt for, and the summary carries both numbers with one sentence on what they would be buying. The same head leads the glance. A brief score of 28 is a correct answer to a question you did not ask, and it used to be the biggest thing on the screen.
This exists because a narrow brief was throwing away good prospects and calling them bad ones. A brief that says law firms in Toronto scores a bank at 8, and 8 is the correct answer to the question it was asked. It is the wrong answer to the question you actually have, especially when that bank is on a competitor's own client list, which is the strongest evidence there is that a company buys this work: they already bought it, from your rival.
One press widens the brief, and only a press does. Where the readings disagree, the studio names the kind of company it found, such as banks and financial firms in the same city, and offers Always look for these. That writes one line to what you hunt for, every later search reads it, and it comes off again on Ideal customer. Nothing is added without the press.
Three things it deliberately does not do. It does not change your leads: the shelf stays governed by your brief until you have widened it yourself. It is not a softer score, so a company that would not buy scores low on both. And it never reaches past a no: anyone on your suppress list stays suppressed, and a rule you wrote about who you will never approach still decides first.
Ask for the leads you want
Describe them and press Find them. Under the leads on your home screen there is a box: "mid-market property managers in the Southeast running class A and B units", "agencies that already resell Shopify apps", "anyone hiring a Head of Partnerships this quarter". From the moment you press, the line under the box says what it is looking for and that the answer lands above. The companies land on the shelf above, with the same Glance press every other lead has.
Or paste a page that lists them. An investor's portfolio, a conference's exhibitor list, a partner directory: paste the address into the same box, on its own, and press Find them. The studio opens that page, takes the companies it names, scores the strongest of them against your brief and puts up to eight on the shelf, each card saying which page it came from. Nothing the page does not name is added, and a page the studio cannot open adds nothing. A page read counts as one of the day's lead searches.
It is the fastest way to two different things:
The studio has your market slightly wrong. Rather than correcting it one company at a time, say what you actually want and get it.
You want to try something new. A market you have never worked in, a one-off idea, a client who asked for something odd. Type it, look at what comes back, and nothing about your workspace changes.
Asking never changes your brief. That is deliberate: most of what people type here is an experiment, and an experiment that quietly rewrote what the studio hunts would make the box risky to use for the thing it is best at.
Ask who moved, not only who fits. "Companies that just raised a Series A", "anyone who opened a UK office this summer", "agencies that changed their head of growth recently", a request that names a time is answered with companies that moved in one. A card that found something says so on one line, "Raised a $40m Series B · 14 Aug", and that line opens the page it was read on. Most companies have not moved and their cards say nothing about when; a fit score is about whether a company is right, and this is about whether today is the day.
We never guess a date. A move is on the card only when the search read the event and the day it happened on a page, with the page linked. Nothing older than six months counts as news.
Prospekt a company that moved and the move writes the email. The run is handed what happened and the day it happened, and inside two months that is the opening line: why this note is arriving today, in the reader's terms. Older than that it argues the case and never opens on it: a note congratulating somebody on something months behind them reads worse than one that never mentions it. The run checks the move still holds while it reads their pages, and drops it if what it finds says otherwise.
A company you typed in gets the same treatment. The full run looks for a dated move itself while it reads their pages, so a company you found somewhere else (Research prospekts, a portfolio list, a finished pitch's lookalikes) can still open on why today. Same bar: read on a page, dated, linked, inside six months; where the run already has a move, the fresher of the two wins.
Build an agent from a description
/agents: say what you want done, and the studio builds the thing that does it. One sentence: "find agencies in Berlin that already run paid search", "read Signaliz and tell me who they sell to", "who should I be calling at Northline?" The studio picks the door (find leads, lookalikes, a glance, a pitch, decision-makers, a competitor report) and fills in what that door needs from your brief.
The agent is kept. Next time it is one press, and the list shows when each one last ran.
Watch a competitor, and it tells you what to do about each move. "Watch Signaliz every week" builds a standing watch: the card lists what that company publicly did since the last look (a launch, a price change, a funding round, a partnership), each with its date, a link, and one line on what to do about it. Nothing public since the last watch is a real answer and the card says so. It reports what a company did, never what a person did.
Say how often and it runs itself. "Check Signaliz every Monday" is a standing agent: the card shows when it next runs, and Stop ends it. Three agents per workspace may stand at once, and a schedule needs a subscription; a free workspace keeps the words and the card says so. To add a schedule to an agent that has none, describe it again with the cadence in the sentence.
It never invents what you did not say. "Watch our biggest rival" names no rival, so the card asks which company and nothing else.
And it says when it cannot. A sentence the studio has no verb for is answered as one, not routed to the nearest thing. The agent is kept anyway, so what you asked for is on the record.
There is nothing else to configure. No model to choose, no tools to switch on, no prompt to edit: the brief you already wrote is the configuration, which is why one sentence is enough.
When it was not an experiment, keep it. After the results land, one press, Always look for these, puts what you typed on your brief, and every later search reads it. They are listed on Your brief under What we keep looking for, and a cross removes any of them.
What it will not do: get past your suppress list, your disqualifiers, or companies you have already been shown. Asking for something new widens where the studio looks; it never overrides who you have said no to.
Is the score wrong? Tell the studio why
Press the score. If a company is a better fit than the number says, the score itself opens one question: what did we get wrong? Type a sentence and press Re-score.
Two things happen, and the second one is the point.
This company is scored again, in about fifteen seconds, with what you said taken as given. The new number is earned rather than typed: you do not set the score, you correct the thing the score was wrong about.
And the studio learns the part that was about you. Most corrections contain two things at once: "Bozzuto is an excellent fit; we serve class A and B units and luxury operators, we are not limited to lower-income housing." The first half is about that one company and stays on its record, where the full pitch and the call sheet read it as your call. If a pitch still scores against it, the pitch's checks say so. The second half is about you, and it goes on your brief, where every future glance, every company the studio suggests, and every pitch it writes will read it.
That is why a wrong score is worth thirty seconds. Fixing the number helps one lead; correcting the premise changes what the studio goes looking for.
A green line tells you exactly what was learned, and Undo takes it straight back off if it is not what you meant. Your corrections are listed on Your brief, where you can edit or delete any of them.
Or say they are a competitor. If the company competes with you, say so in the same box. They are not scored again, because how good a prospect they are is the wrong question about a company you sell against. They go on your rival list instead, so the studio stops offering them and hunts the market around them. Undo takes that back, and they are listed on Do not contact.
What makes a good correction: a fact about your company that would still be useful for a company nobody has mentioned yet. "We sell to CFOs, not office managers" teaches the studio something. "This one is good" only fixes this one.
Name the people: one press on the glance
Name the people sits under the roles on the glance card. One press buys the searches the glance would not spend: about a minute, capped at four pages and four searches, and the roles become named people, each with the source that supports the name, and a LinkedIn link only where a profile was actually read. A role nobody could be sourced for is then asked of the contact vendors, who answer with the person who holds it and their profile; where they have nobody either, the role stays a role, with a null name, because a guessed name is worse than none. A named person without a profile link is looked up the same way, so the badge beside the name is their profile wherever a vendor knows it. People on your workspace's suppress list are never returned, and a company on it refuses the press the way it refuses a run. The card prints the timing against the sixty-second ceiling, the way the glance prints its fifteen.
What to send: one piece at a time, or everything
Under the glance you decide what to buy. What to send offers each piece (pitch deck, one-pager, cold email, proposal) as its own press: one call, under 90 seconds, written in your voice with everything your workspace has taught the studio, and written to read as a person wrote it: no dashes, no stock phrases, plain verbs. Pressing one opens the pitch straight away (the glance's score and argument are already on it) and the piece lands on its own tab while you watch. Each further piece is another press on its tab; the pitch grows as you buy, is saved to your library like any other, and every piece is editable and shareable the moment it lands. A piece a pitch does not have yet always shows its press, on staged pitches and on full runs where you unticked it alike. The compliance check runs on the whole pitch after every press.
A glance itself is not saved to your library and does not use up anything. Press Prospekt → on the card and the six-step run starts on the same company: deeper research, everything written at once.
Starting a full run
Research prospekts takes a company name, a bare domain, or both. Behind More options, all optional: notes, a point of contact, which outputs to write, the motion to argue for and the value angle to lead with (the last two come from your workspace's own brief). Notes are read as your word on the company: where they say something the brief does not, they win, and the score follows them. Until you have edited Your brief yourself, the studio treats what it drafted there as its own reading rather than your rule, so a line it wrote about who is too large or too small can lower a score but never disqualify a company on its own. The motion on a finished pitch is one your brief lists, in your own words. A brief that lists none is scored as a direct case (every company judged as a buyer of what you sell) and the studio never argues a reseller, affiliate or partner route the brief does not name; where that is the only case, it says so and the score stays low. The same happens when your brief lists motions your goal does not support: a goal that hunts companies to sell to, over a list of partnership routes, is argued as the goal and the pitch comes back with no motion set. If you see that, the list is stale. Edit it on Your brief, or rescan.
Once you have named a company, the Prospekt button says how long one takes: Prospekt · ~7 min. The number is the middle of your own workspace's finished runs, so it reflects your brief, your deliverables and how much reading your targets need. It appears once three runs have finished; before that we do not know yet, and would rather say nothing than guess. Runs are included in your plan; the minutes are the only thing a run costs you.
Before anything else, the studio works out who the company actually is from its own website: it opens the site, reads what they sell and to whom, and rules out same-named businesses. The pitch is then researched, scored and drafted against that settled identity.
While it runs
A run takes a few minutes (the button that started it said how many, measured off your own finished runs) and shows six named stages: understand, research, score, draft, people, check, with what each stage is actually doing ("2 searches · 9 sources read"; the counts are that step's own work). The stages are not strictly single-file: the decision-maker search needs only the confirmed target, so it runs alongside the drafting rather than after it; both steps light up at once while that happens, the foot names both ("Steps 2 & 5 of 6 · BDR + SDR agents"), and step 5 ticks done the moment its list is ready, even while the drafts are still being written. The drafts themselves are written side by side: each deliverable is its own call, the same call its press on a finished pitch makes, so the drafting step takes as long as its slowest piece, not the sum of all of them. A piece that fails does not fail the run; the finished pitch ships without it, and its tab holds the press to write it. There is no percentage bar on purpose: research is one stage of six, and a long read would sit at 17% looking dead. The clock counts from when the run started, not from when you opened the screen, so coming back to a run four minutes in says four minutes. When the draft lands, the terminal ends with where the time went: each step's own clock, from the run's own record, including what ran in parallel.
A company the studio already researched is not researched again. Research on a target is kept for a month and shared across every workspace on the deploy, so a rerun (or another workspace pitching the same company) starts from the settled identity, and the more Prospektor is used the more of your research is already done. The run says so ("Already researched today: …, read once for every workspace on Prospektor"), and it caps its remaining reading to a small, named freshness budget ("Research already settled: capped at 3 pages · 2 searches"): recent news and anything time-sensitive, never a re-read of pages it already holds. What crosses is the reading of a public website and nothing else; the box carries nothing about the workspace that paid for it, and nothing of your brief, your client or your edits ever reaches it.
The steps on the left keep what they found. The terminal on the right streams the run as it happens and scrolls; the steps beside it fold each finished step into a summary that stays for the rest of the run: what your brief brought to it, who the company turned out to be and which same-named businesses were ruled out, every page it opened (each one a link you can open and check) and every search it ran, then each deliverable as it is written. By the last step that column is the run's own account of itself, and you can read it without having watched.
Under the six, greyed out, are the three things that come after a finished pitch: send or share it, prep the call it books, record what happened. They are signposts to screens you already have, not steps this run takes.
Each step says who is doing it. The right-hand column of the rail names the agent working on that step, and the foot of the run names the one working now. It is a GTM team, and these are the five:
- BDR agent brings you companies worth pitching. Opens each target's own website before anything else and works out who they actually are, not who shares their name; scores fit against your brief; suggests the next ones worth your time and names the ones it ruled out. (research, score)
- SDR agent finds you the way in. Names the decision-makers with a source behind every name and a confidence rating on each, and leaves a blank rather than inventing one; then finds the warm path through your network. (people) Companies share names, so a name is checked against the company the BDR agent settled on before it reaches you: someone a source places at a same-named company elsewhere is left off, and confirmed is only given where the source ties the person to this company's own site; the rest read likely. Fewer names, every one of them at the right company. And when the search comes back empty but the research already read profile pages naming people here, those people are on the list, with a short second look to confirm their roles.
- AE agent writes what you send. Drafts the deck, the one-pager, the proposal and the email sequence in your voice, grounded in what has already worked for you; preps the call it books. When somebody at the company reacted to or commented on one of your posts (pasted under Your network), the opener says what they reacted to. (draft, and prep afterwards)
- Sales Ops agent checks the work before you see it. Reads the finished draft back against the rules and flags anything that broke one: an invented number, a fabricated person, a banned word, a company the research already ruled out. It never rewrites your copy; it makes the problem visible next to the thing that has it. (check)
- RevOps agent remembers what worked. Every outcome you record and every edit you save goes into the next run, so run 100 is not run 1. (learn)
Two steps carry no agent, and both are deliberate. Understand the request is your own brief read back; you wrote it, so nothing else takes credit for it. And send or share it says You: the agents research and draft, and you send. Nothing here has your relationships, and nothing signs your name.
The rail's order is the run's order. A run drafts the deliverables before it names the decision-makers, so the AE agent appears above the SDR agent on the rail even though a sales team would introduce them the other way round. That is the true order the work happens in, and the rail shows it rather than a tidier one.
Setting up a workspace shows a shorter, three-step rail with no agents on it. That job is the studio reading your site to draft your brief, not a pitch, and it should never look like one.
You can leave. The worker runs in the background; Running lists every run in flight with Watch to drop back in. A run nothing has been heard from in twenty minutes is marked stopped and offers Dismiss; a run still working offers Stop (confirmed, because it kills real work). There is no pause.
Busy moments are retried, and a failure is a fresh start. The model provider occasionally has a busy moment; the run says so in its log, backs off and retries by itself; nothing you entered is lost. A run that fails anyway says why in one plain sentence and offers two buttons, there and in Did not finish: Rerun starts a fresh attempt at the same target with the same ticked deliverables (a glance that failed reruns as a glance, never as the full run), and Share this bug with Prospektor sends that run's error and step log to the Prospektor team for analysis, only when you press it, never on its own.
Everybody reads every run; your own are yours to manage. Stopping, archiving or deleting a run somebody else started, or calling off their batch, needs the owner or admin role in the workspace. Reading it never does. (The roles are in Workspace settings.)
Beast Mode: a whole list at once
The foot of the Research prospekts form: upload a CSV or paste one company per line (Company name, website, the website optional; with a header row the columns are read by name, so a CRM export works as it is) and Run batch → queues every row through the same pipeline. Up to 100 companies per batch, three running at once, each landing in the Library like any other pitch. The CSV is parsed in your browser and never uploaded. A free workspace runs its one full pitch from the form above, one company at a time; a batch is the rest of your list, and pressing it opens Upgrade.
One email at the end, not one per run. Close the tab; when the last target lands, the studio emails whoever started the batch: which companies scored 75% or higher, named, with their own scores and the angle each run chose, a link straight to each of them, and how many runs landed out of how many you queued. If nothing cleared 75% the mail says so and names the closest instead, because that is also the answer. Fifty targets are never fifty emails, and it is off in one click from the mail itself or in Workspace settings.
Searches: the Library's rows
Every company in the Library got there through a search: a lookalike search, an ask typed into the lead box, a page you pasted, a list you uploaded, a rival's client list from a Counterprospekt report. The Library lists those searches, one row each, and the cards live inside. Press a row to open it; inside, the cards sit strongest fit first.
The studio names each search from what you asked. Like SplitPay for a lookalike search, your ask as you typed it, the page's own name, the name you gave an uploaded list or its day (Batch, 29 Sep), Rival Inc's clients for a report. Two rows are standing: Found for you holds what the shelf found on its own, and One at a time holds companies typed on Research or sent by a machine token.
Rename a search with one press. The pencil beside the name turns it into a box; Enter or Save keeps it, Escape puts the old name back. Names are free text, shared with everyone in the workspace, editable by any of them, never sent anywhere, and kept with the workspace.
Move a company with ⋯ → Move to… on its card, for the one time a company belongs in a different search than the one that found it. New search… at the foot of that list makes one and files the company under it.
A dot before a row's name means something in it is unread. One search is no list: while the Library holds a single search, its cards are the screen. The Running strip above the rows says an uploaded list's own name while it runs.
Glances land the same way. A glance lands in the Library beside your pitches, marked Glance, under the search that found the company.
Reading the result
A pitch reads as a sequence: Summary → Fit thesis → Signals (for and against, each rated by recency and recurrence) → Deliverables (deck, one-pager, cold email, proposal, each block copyable and pencil-editable) → Decision-makers → Lookalikes. If it is coral, you can click it.
A run writes the cold email, and leaves the rest to you. The deck, the one-pager and the proposal each wait on their own tab with a press button, one call, under 90 seconds, whenever a pitch turns out to want one. Nothing is missing and nothing is slower to get; the run simply stops writing three documents in advance for every company you look at. Decision-makers always come with the run, because the research call writes them as it goes. Lookalikes come with a run that scores 50 or better: a lookalike is a company that fits for the same reasons this one does, so a weak or not-a-fit target has none, and its tab says so rather than listing the target's competitors.
A lookalike row is a name, a similarity and one line on the resemblance; never a profile. The run researched the target, not them, so nothing on the row about the company itself comes from memory; where the name was read (the target's own comparison page, a competitor list) the row links that page in one word. Press Glance on the row to read the company itself.
The Summary is one look, on purpose: the score and name up top, how big they are, and The one thing, the single sentence the pitch rests on. Everything else the studio worked out (the firm in a paragraph, the four numbers, what the run drew on from your workspace, and the identity check) sits behind The full picture, one press away on the same tab.
The Fit thesis reads the same way: each bullet of the case and each risk is one line you can scan, and the objection you will hear comes with a counter of a few complete sentences you could say back verbatim. Every line ends properly: the studio rewrites for shortness rather than clipping, so nothing trails off mid-sentence. The prose version, the argument in full, sits behind The full argument, one press away on the same tab. The other tabs carry the full detail as they always have.
The first deck asks one question. The first time anyone in the workspace presses Pitch deck on a run, the tab asks what your decks should be built from, with three answers. Write a standing deck drafts one from your brief in about half a minute and then builds this company's deck from it. Upload a reference does the same from a deck you already send: pick the .pptx (or a .txt or .md), the studio reads the slides' text in your browser and keeps the order and register, and the file itself never leaves your machine. No template writes this deck on its own, one call, and every deck after it the same way. Whatever you answer is remembered, so the question is asked once; Workspace settings → Deliverables → Standing deck is where the answer is changed. Only an admin or the owner is asked, because the answer changes every deck the workspace sends.
The deck is your standing deck, plus one slide. Once the workspace has a standing deck, a pitch does not write one: it reuses those slides, drops in a slide about this company right after the opening one, and fills your prospect's name and business into the gaps the template left. That is instant and costs no call. A workspace that answered No template still gets a deck the old way, pressed on its tab.
Three deliverables carry their own shape. The pitch deck is real slides: Download .pptx on the deck tab writes a 16:9 file that opens in Keynote, PowerPoint and Google Slides, one slide per block, in your workspace's own colours, with the speaker notes on each notes page and the figures' sources folded in beside them. It is built in your browser, so nothing about the deck leaves the tab. Edit a slide with its pencil first and the download carries the edit. Its last slide ends on one quiet line, Researched by Prospektor, the run's source count and the date it ran, the same mark a shared pitch carries, on the file that gets forwarded furthest. There is no switch to remove it. The one-pager is a value-driven, print-ready proposal page: your logo and name up top, why this matters to the reader before why you are the right one to do it, proof in one-line points, and one unmissable call to action; Ctrl/Cmd-P turns it into the page it already looks like. The proposal is the formal document you send once they are interested (scope, commercials and timeline, ready to forward) and its tab says so, so nobody wonders when to use it.
- A name appears only when a source supports it: a page the studio read, or a contact vendor's record of who holds the role. A role nobody could be named for is asked of the vendors, and stays blank rather than guessed when they have nobody at the company in that role. A named person without a profile link is looked up the same way; an older pitch gets its profiles the next time it is opened.
- Work emails arrive on their own. Every named person is looked up with the contact vendors the studio has keys for, cheapest first, then a routing service that tries eight more, and the address lands beside the LinkedIn badge, with what verified it in the tooltip; a catch-all domain is said there too. Where Moltsets found it, the tooltip also carries its A to F grade for the mailbox (A validated, B a past send landed, C catch-all, F no signal; an address it grades D is never shown). No address means no vendor had a verified one at the company's own domain; an address at a company the person has left is dropped, and so is a shared desk like
info@. Emails are included without a cap; the CSV download carries them, a shared link does not. A pitch from before addresses existed fills in the first time you open it; give it a few seconds. - Phones are one press each, and counted. A dashed phone? beside a person's address asks the vendor for their number, by the verified work email first, so it is theirs at this company. A number found lands there as a tap-to-call link and takes one of your workspace's 25 phone lookups for the month; the count left is in the button's tooltip. Nothing found costs nothing and reads no phone. The CSV carries numbers; a shared link never does. No phone? at all means the free plan, or a studio whose lookups are switched off.
- Bracketed placeholders in drafts mark what could not be verified. They are marked wherever they appear, and the pencil on any block is where you fill one in, which also teaches the studio your wording.
- Find more under the researched lists queues more research in the background; an empty box is a valid instruction ("look again").
- One specific person goes in the same box under Decision-makers: a name (Jane Doe), a LinkedIn profile URL, or either with a role beside it (Jane Doe, CFO). That person joins the list as you typed them and their work email and profile are looked up, the way a run's people are. Someone already listed is not added twice. A role on its own (head of customer operations) still searches for more people.
- Compliance findings: the studio still reads its own draft back for banned terms and numbers without provenance, and says what it found in the run log and the run-ready email. It is a check on the draft, not a screen to work through. When the studio could not read the company's own site, a finding says so in the studio's own words (came back empty for a site drawn in the browser, the check did not complete, or could not open) and counts the pages the draft read for itself. That is a note about the studio's reading, never a fact about the company: a site the studio cannot draw still has its pricing page, and the pitch does not score the empty pages as evidence of anything.
- The pages are numbers on the one thing. Every page the run read is a small superscript at the end of The one thing on the Summary, one number per page. Hover a number and it says what that page established; press it and the page opens. There is no list of sources anywhere else on the screen, and the sources read count under the summary counts exactly those numbers.
- Who this is, behind The full picture on the Summary, is the studio's answer to the question everything else rests on, in one line: "Identified as Matte PR by reading 11 of their own pages · 3 similar companies ruled out." Under it, and open, are the same-named companies it checked and decided were not your target, each with the reason. If a company you have never heard of is missing from the ruled-out list and its facts are in the profile instead, that is the thing to catch before you send anything.
- Sometimes it says "read 12 days ago", and that is not a shortcut being hidden from you. Working out who a company is means opening their site and reading it, and the answer does not change from one week to the next, so once any workspace on Prospektor has established it, the reading is kept and handed to the next run against that domain instead of being bought again. That is why a run can reach the same conclusions in seconds. When the reading is not this run's own, the line says when it was made, and the pages and ruled-out companies behind it are the ones read that day. Nothing older than a month is reused; past that the site is read again from scratch. If you want a fresh reading sooner than that, say so to support and we will clear it for your target.
- The studio stops looking once it is sure. Confirming who a company is takes a handful of pages, not a hundred: it opens the site, reads the page that names the operator, checks that nobody else trades under the name, and stops. It spends the full check only where the answer is genuinely in doubt (the site would not open, the name you typed is not the name on the site, something claims the company has stopped trading, or another business is still in the frame). That is why the line above sometimes says four pages and sometimes eleven: it is what the answer actually took, not a quota.
- The highlighted sentences in a draft were written for this company. Above each deliverable a line counts them ("6 of 21 sentences were written for Delvinia") and the sentences themselves are marked where you read them. A sentence counts when it names something the run established about the target: its name, brand, legal entity or domain, one of the decision-makers the research named, where it is or how big it is. Its industry does not count, because B2B SaaS describes the target and everyone it competes with. Unmarked is not wrong: your own proof and your own offer belong in the draft and are not about the reader. The one to act on is a draft that names nobody: the studio says so in red, and it means the draft reads as something you could send to anyone. On a shared link the recipient sees the count and the highlighting, and never that judgement.
- From your own workspace, behind The full picture on the Summary, says what this run had of yours to read: how many recorded outcomes and how many saved edits went into the draft. Everything else came from your brief and the live web. A run that found no outcomes says so, and offers the door to record one; that is the loop that makes a later run better than this one.
Editing, and what it teaches
The pencil on any deliverable block edits it in place. Saved edits become brand-voice examples for future runs: tone and structure only, never facts, and facts never cross companies. Nothing else a team does in the studio teaches it as much.
Since 23 Aug 2026 a run chooses those examples for the company it is working on rather than simply taking the three most recent: the ones whose subject looks most like this target, with at least one example of each deliverable the run has to write. So a workspace with fifty saved edits draws on all fifty, a few at a time, instead of on whichever three were saved last.
Since 25 Aug 2026 the save also records how long the block was open before you pressed it. That is the studio measuring itself, not you: a draft you had to spend four minutes fixing was a worse draft than one you fixed in twenty seconds, and that is the only way this product can honestly tell whether a change to how it writes actually helped. The record is a duration and the kind of deliverable, with no address on it and nothing you typed; only Prospektor's operator sees it, as a median across many runs, and nobody in your workspace or the agency above it ever does. Privacy says it in full, and a browser sending Global Privacy Control switches it off while your edit saves exactly as it would have.
A run also carries what your team recorded in Outcomes: what actually won and lost, in their words. The waiting line says so while the run works: "Brief loaded, with 4 of your edits as voice references and what you recorded about 6 companies." It shapes the judgment, never the copy; a company from your own outcomes record appearing in a deliverable is flagged by the compliance check like a banned word is.
Ask the studio
The last tab of a finished pitch is a conversation about it: where a claim came from, how to open the first call, what to say when the predicted objection lands. It answers only from that pitch and your brief, and never changes the pitch. Each member has their own conversation per pitch; it survives reloads.
Questions come out of a daily allowance that every typed-sentence door shares: this tab, the support chat, the home box, the agent builder and Make sense of this. It is roomy on a paid workspace and smaller on a free one, and when it is spent the studio says so in one line and answers again tomorrow.
Print / PDF
Ctrl/Cmd-P prints the deliverable styled with your workspace's own stylesheet (Save as PDF in the same dialog is the PDF export). A deck prints as a deck: one slide to a landscape sheet. For a file you can present from and edit, use Download .pptx on the deck tab instead.
There is no Print / PDF button: it duplicated a keystroke your browser already has, on a screen that reads better without it.