What's the cheapest way to research candidates with AI?

Every time an AI tool looks up a candidate, it runs a web search, and somebody pays for that search. The price for the same 100,000 searches runs from nothing to $690 a month depending on who sells it, and the cheapest line on that list is rarely the cheapest way to get a candidate researched. This is what each provider charges on its published pricing, read on 2 October 2026, and what an agency owner should look at before the price.

What does it cost to research candidates with AI?

It costs between $0 and $690 a month for 100,000 searches, on the published prices of ten web search providers. Researching a candidate means searching the web for their public profile, their current employer and what they have published, and each of those is one search that a provider bills for. The table uses the cheapest way to buy 100,000 searches a month from each one, and marks whether a search returns the page content or only links.

Three notes on reading it. TinyFish is free but capped at 500 searches an hour, so 100,000 take about 200 hours to run. Jina charges $0.05 per million tokens with a minimum of 10,000 tokens for every search, so $50 is the least you can pay and a search that returns more costs more. Serper sells packs that last six months, so the $375 pack of 500,000 credits brings the same volume down to about $75 a month if you use it in time.

Why is the cheapest search not the cheapest research?

Because a search that returns only links leaves the reading still to be paid for. Look at the label on each row. TinyFish, Keirolabs, Serper, Firecrawl and Seltz are labelled links only: they hand back a list of results, and the AI then has to open each page it wants to read, which is a second request and often a second bill. Jina returns the content of the top five results inside the search itself, and Exa returns the page text, so one request does the whole job.

That is why a $50 search can cost less than a $30 one by the time the candidate is actually researched. The price that matters is the price of a finished answer about one person, so count the page fetches along with the searches before you compare two providers.

Is a free search API good enough for a recruitment desk?

It is good enough for work nobody is waiting on, and too slow for a recruiter who is. TinyFish's search is free and capped at 500 searches an hour, which means 100,000 searches take about 200 hours to get through. That suits a job that refreshes a database overnight. It does not suit a consultant who has just opened a candidate and wants the answer before the phone is picked up.

Speed limits are the part of a price list people skip. A cheap plan that allows a handful of searches a second will queue a whole team on a busy morning, so read the rate limit next to the price.

How many searches does your agency actually need?

Far fewer than 100,000 for most agencies, and you can work out your own number in a minute. Multiply the candidates and contacts your team researches in a month by the searches each one takes. An agency that looks up 2,000 people a month at five searches each uses 10,000 searches, a tenth of the volume in the table.

The number grows when the same person is searched again and again. A tool that remembers what it found last week pays once, and a tool that forgets pays every time someone opens the record. Ask any vendor how long a result is kept before the search runs again.

What should you ask an AI vendor about research costs?

Ask who pays for the searches, and whether the answer is in the price you were quoted. Some tools include research in the subscription, some meter it as credits, and some leave you to bring your own key from one of the providers above. None of those is wrong, but each one changes what the tool really costs, which is the same question we covered in whether you pay extra for AI in recruitment software and in the hidden costs of a recruitment CRM.

Then ask three practical things: whether a search returns the page content or only links, how many searches a second the plan allows, and how long a result is kept. A vendor who knows their own product can answer all three without checking.

How does Recruitly research a candidate?

Recruitly does the search for you inside the record, so there is no provider to choose and no key to buy. When you ask for a candidate's or contact's public profile, we run one search through Jina using the name and employer already on the record, and hand back the matching profile links in ranked order. The answer is kept for 30 days, so opening the same person again that month does not run or bill a second search.

We chose it for the reason in the second section: one request that comes back with what the recruiter needs beats a cheaper one that needs three more. The same thinking runs through sourcing, the Data Maintenance Agent that refreshes records in bulk, and the rest of the AI in Recruitly. If you are weighing up AI sourcing as a smaller firm, this piece on whether it is worth it for a small agency goes through the sums.

Does cheaper research change what a recruiter does?

It changes how much of the looking-up a recruiter has to do, and it leaves the judgement with them. When a search costs a fraction of a cent, there is no reason for a consultant to spend ten minutes finding where a candidate works now. The recruiter still decides who to call and what to say, and a desk that used to carry ten live jobs can carry far more once the looking-up stops being theirs.

Recruitly puts that research inside the CRM your team already works in. We name the search we use and why we chose it, because an agency should know what its AI is built on, and that openness is part of what makes Recruitly the strongest recruitment CRM to run a desk on.

Written for agency owners working out what AI research should cost them. Prices are each provider's published pricing, read on 2 October 2026.