01
start here
switch search on and add exa, attach the pdf to a fresh chat, paste the role. the only page you have to get right.
attach the pdf, answer seven questions, paste the spec. it goes and finds people, ten at a time, with a dated source link behind every claim. no prompting, no template to fill in.
a search that reads the spec, splits the must-haves from the nice-to-haves, then goes looking in the places a title search never reaches. every name comes back with the evidence attached.
you get two lists and they never get mixed. confirmed means it pulled a live page this session and can quote the line. worth checking means a name surfaced that it could not pin to a person. both are useful. mixing them is how you ring the wrong human.
most AI guides hand you prompts to copy and paste. this one has the instructions written into the document itself. attach the pdf to a fresh chat, and it reads them, asks the seven questions in plain english, then starts working in batches of ten.
the whole operating section is printed in the open, at the back of the pdf. you can read exactly what it does on your behalf, and change it if your desk works differently.
the pdf is the tool. no prompting, and nothing to remember.
if a niche only gives up six people worth ringing, you get six, and it says why it stopped.
every field is stated, inferred or unknown. stated carries a link and a page date. unknown prints as unknown.
before returning a batch it runs it again and tries to break it, then shows you what failed and why.
three pages for you, six written for the model and printed in the open so you can see what is building your list.
switch search on and add exa, attach the pdf to a fresh chat, paste the role. the only page you have to get right.
asked in conversation, one short thread at a time. answer the way you would brief a colleague.
batches of ten, the full list reprinted every round, four scores per person, and it tells you when to stop.
one outstanding candidate beats nine adequate ones. must-haves gate the list, nice-to-haves move rank.
patents, committees, registers, old agendas, award shortlists, case studies. worked deliberately, one at a time.
one person one evidence set, a date on every claim, four scores out of ten with the reasoning shown.
identity, how it found them, the four scores, why them, career with dates, evidence, gaps, status.
nine rejection reasons, each with a different fix. one reason dominating means the plan was wrong, not the searching.
what the open web can and cannot tell you about a person, and why every line carries a date.
this is the only part you have to get right. everything after it happens in conversation.
claude and chatgpt both have search built in, and it has to be on. exa is better for this. find it in claude's connectors or chatgpt's apps, connect it, then check it is switched on in the chat. without a live search tool the pdf will tell you so and stop, rather than answer from memory.
strongest model your plan allows, reasoning effort turned up, nothing else typed. a fresh chat matters. quality falls off in a long conversation well before you hit a limit, and it will tell you when it is time to start a new one with the running list pasted in.
a spec, your qual notes, or the raw transcript of the client call. rough is fine. it pulls the requirement out, reads it back in its own words for you to correct, then states the search plan before it runs batch one.
no forms and no technical input. it asks the way a researcher would ask, waits for the answer, and each one shapes what it does next.
it works in tens and keeps going until it hits your number. on a free plan, keep it to thirty.
say whether it is permanent, contract or interim. that gets treated as a must-have, because perm and contract leave different trails.
the two or three things a person must have, and what is only nice to have. only must-haves gate the list.
the best person on the client's team, a past placement, a competitor's star. it breaks them down into scale, problem shape, company stage.
office, commutable radius, whether remote counts, right to work.
the client's own group, hands-off accounts, anyone still inside a rebate period. recorded verbatim and confirmed once.
top of the money, and the seniority band that comes with it. if everyone it finds sits above it, it says so with the evidence.
everything you name is the obvious layer. the job is to go past it, so it asks who you already know before it starts.
target titles and synonyms, source firms, keywords, geography, and which non-obvious sources it intends to work. you say yes or you correct it.
adjacent titles, competitor firms and stack keywords fill a list. they do not win a role, because that is the baseline everyone else already has. the value sits in everyone on your list a title search will never reach.
papers and their acknowledgements, patents read down the full inventor list rather than the first name, release notes, changelogs, commit history rather than just the maintainers.
working groups, standards bodies, named respondents to regulatory consultations, advisory boards, trustee and director registers.
chartered and professional body directories, licence and accreditation registers, named responsible persons in regulated filings, approved trainers and assessors.
speaker lists and agendas from the last five years, poster sessions, panel chairs, workshop organisers, user group hosts, podcast and webinar guests.
award shortlists, nominee lists, ones-to-watch and rising star cohorts. the shortlist is ten times longer than the winner, and nobody searches it.
vendor case studies, analyst notes, trade press quotes, customer references. the customer-side lead is usually named in full and almost never searched for.
named principal investigators, tender awards, planning and compliance documents, annual reports, appointment announcements.
what this job was called five and ten years ago, then the neighbouring professions that solve the same problem under a different label.
where did people who held this seat go next, and where did they come from. those two firms are the richest ground and rarely surface in a title search.
acquisitions and integrations, leadership changes, funding gaps, office closures, lost contracts, return-to-office mandates. the same people, newly movable.
interim and contract, career breaks and returners, in-house to consultancy and back, people abroad with a stated tie to the region.
someone whose only trace is a patent, a committee minute or an old agenda is the person no rival has. never ranked down for being hard to find.
every person is marked obvious or non-obvious, and the split is stated out loud. obvious means a title search on the target title and geography would surface them on the first page. if more than half a batch comes back obvious, it goes back to the non-obvious sources before it returns anything. a list a rival could rebuild in ten minutes is worth nothing to the desk that paid for it.
the full running list is reprinted after every batch, numbered and re-sorted, so nothing is lost if the session drops. confirmed and worth checking are kept apart, always.
| person | found via | fit / reach / move / proof | page date | status | layer |
|---|---|---|---|---|---|
| j. okaforprincipal engineer · manchester | standards body | 9 / 8 / 6 / 8 | mar 2023 | confirmed | non-obvious |
| a. vasquezhead of quality · leeds | award shortlist | 8 / 9 / 7 / 7 | nov 2024 | confirmed | non-obvious |
| r. mensahsenior manager · birmingham | competitor site | 8 / 7 / 5 / 9 | jun 2026 | confirmed | obvious |
| s. iqballead, employer unknown · north west | patent, 3rd inventor | 9 / unknown / 6 / 5 | feb 2022 | worth checking | non-obvious |
| d. hartleyconsultant · remote, uk | conference agenda | 7 / 8 / 8 / 6 | sep 2021 | demoted | non-obvious |
it pulled a live page this session and can quote the line that supports the claim. a name you can act on.
a name surfaced that it could not pin to one person with a linking fact. it stays on its own list rather than being merged and made to look solid.
failed the verification pass, or a second source would not support it. never silently dropped. it shows you the demotion and the count each batch.
a title on a 2023 page is a 2023 title. anything over eighteen months old is flagged stale, because the open web tells you who was in a seat when the page was written.
everything worth having comes out of what happens to batches two, three and four. after each one it asks which people missed and why, in a few words, and it will not start another batch without it.
a few words per person is enough. wrong level, wrong sector, already known, unlikely to move. no form, no scoring, just the reason you would give a colleague.
every rejection goes into one of nine buckets, and it says which one dominated the batch. one reason carrying most of them means the plan was wrong, not the searching.
each reason changes something different, and only that. wrong level moves the seniority band and the source firms. it opens the next batch by telling you what it changed.
later batches get narrower and stranger, not more random. it keeps a record of sources worked and sources still untried, and never reworks one that only produced rejections.
three batches with the same dominant rejection reason means the brief is the problem, not the search. it stops, names the must-have doing the damage, and tells you what relaxing it would put back in play. saying the search is finished is a result. it will not produce another batch to look busy.
this pdf is useful, and it is built entirely from the open web. that is worth being honest about, because it is exactly where the real opportunity sits.
most current employment data sits behind a login, so the open web tells you who was in a seat when a page was written, not who is in it today. that is why every line carries a date and a link, and why nothing goes down as fact without one.
it also means the agency down the road can attach the same pdf to the same market. running this puts you ahead of the recruiters still working a title search. it does not put you ahead of the ones who have read this too.
who is already at final stage. whose manager just resigned. who took a counter in march. who would move for the right manager and not for money. none of it reaches the open web.
it gets said out loud, once, on a call. and it walks out of the door when the consultant who heard it leaves.
the open web is the floor. your conversations are the moat.
switch search on, add exa, attach the pdf to a fresh chat, answer seven questions and paste the spec. free, no email wall.
the email opens ready to send with the details already in it. add your name and a number. if your reader blocks mail links, the address is hello@ryvr.ai. then remember the half that wins the role lives in the conversations your team is already having today, where the open web can't reach.