Indeed
recommendation engine that becomes the main source of conversion on a marketplace, ahead of keyword search
On May 15, 2026, Hisayuki Idekoba, CEO of Recruit Holdings, said that 70% of applications on Indeed now come from recommendations and AI tools, against 30% for keyword search; monthly active users hit a record in March 2026, up 18% year over year, employers using Sponsored Jobs respond 45% faster to candidates, and Indeed claims 31 hires per minute, up from 27.
Key points
- Indeed has shifted the application from keyword search to AI recommendation.
- The setup relies on Sponsored Jobs, Smart Sourcing, Smart Screening, and Career Scout.
- 70% of applications come from recommendations, monthly actives up 18% in March 2026, 31 hires per minute.
- Figures stated by the CEO at the annual results and repeated in an Indeed press release.
Objective
Make recommendation, rather than keyword search, the mechanism that brings a candidate to apply. Two things are at stake: giving time-pressed employers applications they can act on right away, and keeping candidates on Indeed while conversational assistants and AI-generated search summaries divert part of job boards' traffic.
The deployment
Indeed is the job platform of Recruit Holdings. Its May 15, 2026 press release gives the scale: more than 645 million candidate profiles, more than 3.3 million employers, more than 60 countries, and integrations with more than 350 applicant tracking systems. The setup addresses one precise point in the journey: the path by which a candidate reaches the job they apply to. The historical mechanism was keyword search. Someone types "French chef", sponsored listings surface in the results, they apply. On May 15, 2026, presenting the results for the fiscal year ended March 2026, Hisayuki Idekoba put a number on the split between the two sources of applications for the first time: keyword search now brings only 30% of them, the remaining 70% come from recommendations and AI tools. What the candidate sees is jobs offered in their feed, emails and pop-ups saying, in substance, that this company seems to want them to apply. Those messages are triggered by what employers do in their own tools: Smart Sourcing surfaces likely-fit candidates to the recruiter, Smart Screening filters applications against criteria the recruiter has set. The hiring-intent signal produced this way flows back to the candidate side. Career Scout and personalized feeds round out the setup on job discovery from skills and stated preferences. Indeed also announced an integration with ChatGPT in February 2026. Two results are put forward on the employer side: an average 20% drop in time to hire for those using Smart Screening in early US tests, and up to 50% with Premium Sponsored Jobs. These are recruiter productivity gains. The marketplace effect reads elsewhere: employers using Sponsored Jobs respond 45% faster to candidates, monthly active users hit a record in March 2026, up 18% year over year, and Indeed counts 31 hires per minute, against 27 previously. The exact country-by-country scope of the rollout is not published.
Results Proof A
Three of the four figures are spoken by Hisayuki Idekoba, CEO of Recruit Holdings, during the fourth quarter and full year FY2025 earnings presentation on May 15, 2026, and appear word for word in the official transcript published by the company on its investor page (S1). The fourth comes from the press release Indeed published the same day (S2). The established HR trade press covers the case and the 70% figure (S3), and the setup is still discussed at the quarterly results of August 7, 2026 (S4). One reservation to keep in mind: all these sources trace back to the company itself, and no third party verifies the measurements. The 31 hires per minute and the 20% reduction in time to hire are explicitly Indeed internal data. And the causal link between AI and the rise in active users is the CEO's reading, not a test result. The company figures for the quarter, US revenue and earnings per share, are not results of this setup and are out of scope here.
How it works
Inferred typical approachThe internal detail is not public. Here is a proven approach that leads to the same result, to adapt to your stack.
The stack in detail
- outil Smart Sourcing Employer tool that proactively surfaces candidates likely to fit the role, so the recruiter can start the conversation earlier. The signals it produces also feed the recommendations sent to candidates.
- outil Smart Screening Pre-qualification of applications against criteria defined by the employer, ahead of human review. Indeed reports an average 20% reduction in time to hire in early US tests.
- outil Career Scout On the candidate side, helps discover roles from skills, career history, and stated preferences, alongside personalized feeds and alerts.
- plateforme Premium Sponsored Jobs Premium version of the sponsored listing, combining distribution, employer brand, and sourcing. Indeed reports time to hire reduced by up to 50% compared with non-sponsored jobs, on US data.
How it runs, concretely
For ops teams-
1Collect the intent signal on the employer side employer / recruiter
What the recruiter does in their tools, filtering criteria, candidates contacted, jobs they pay more for, indicates which companies genuinely want to hire and for which profiles.
-
2Match profiles and roles AI
The engine crosses skills, experience, stated preferences, and employer activity to identify the pairs where both sides have an interest in saying yes.
-
3Go to the candidate instead of waiting for their search AI
The job is pushed into the personalized feed, by email or by pop-up, with a message saying the company is looking for this profile. That path now brings in the bulk of applications.
-
4Pre-qualify and answer fast employer / recruiter, assisted by AI
Smart Screening filters out off-criteria applications upfront so the recruiter spends their time on the rest. The hiring decision stays human.
-
5Loop back on the actual hire data team / AI
The outcomes reported through the applicant tracking system integrations come back to feed the matching. A completed hire, a candidate who has become unavailable, a filled role change the following recommendations.
The actual hiring outcome, not the click or the application. Indeed says it can see what happens after the application thanks to integrations with more than 350 applicant tracking systems and to small businesses that run their entire hiring on the platform. Without that hiring feedback, the engine falls back on surface signals, application volume replaces quality, and the problem described by the CEO returns: the candidate applies and nobody answers.
How your customers perceive this type of use
Sourced studiesLe paradoxe est documente des deux cotes : 71% des consommateurs attendent des interactions personnalisees et 76% sont frustres quand elles manquent (McKinsey, 2021), mais 75% declarent ne pas acheter aupres d'organisations auxquelles ils ne confient pas leurs donnees (Cisco, 2024). La « creepy line » est localisee : messages recus quelques secondes apres une recherche et suivi de localisation sont les pratiques qui mettent le plus mal a l'aise (Periscope by McKinsey, 2019).
Acceptance conditions
- La confiance dans le traitement des donnees precede l'achat : 75% ne achetent pas sans elle (Cisco 2024)
- Un cadre legal protecteur rassure : 59% des consommateurs disent que des lois fortes sur la vie privee les rendent plus a l'aise pour partager des informations dans des applications IA (Cisco 2024)
- La personnalisation elle-meme est attendue quand elle est consentie : environ la moitie des consommateurs (US 55%, UK 52%) disent s'inscrire souvent ou parfois a des services personnalises (Periscope by McKinsey 2019)
Red lines
- Le message declenche quelques secondes apres une recherche ou un achat : deuxieme ou troisieme cause de malaise selon les pays (Periscope by McKinsey 2019)
- Le suivi de localisation percu comme de la surveillance : 40% de malaise en Allemagne et au Royaume-Uni (Periscope by McKinsey 2019)
- Le mesusage des donnees personnelles par l'IA, devenu la premiere inquietude des consommateurs, a 53% et en hausse (Qualtrics 2025)
Sources: McKinsey & Company 2021 · Periscope by McKinsey 2019 · Cisco 2024 · Qualtrics 2025
How to replicate
Inference, not sourcedData prerequisites
- A demand-side profile rich enough to recommend without a query: skills, history, stated preferences
- An intent signal on the supply side, that is, what the seller or employer actually does in their own tools, and not only what they declare
- Reporting of the final outcome, here the hire, without which the engine can only learn from the click
- Outbound channels wired to the engine: personalized feed, email, notification, with the consent management that comes with them
Org prerequisites
- Accepting that search stops being the main path, which moves product work from the relevance of results to the quality of outbound messages
- A shared definition of the conversion that matters: here the hire, not the application
- A legal framework for automated pre-qualification, mandatory as soon as the decision affects access to employment or credit
- A feedback loop with the tools on the client side, without which the outcome stays invisible
Possible stack
- In-house recommendation engine on a data warehouse, with an online scoring service
- Managed recommendation components to start, before bringing it in-house
- Email and push platform wired to the engine, not to static segments
- Connectors to the client-side business tools to retrieve the actual outcome
The plan, step by step
- Step 1Instrument the journey so that for each conversion you know which path the user came through: search, feed, email, notification.Deliverable: A quantified split of conversions by path, used as the baseline.
- Step 2Define the conversion that counts from the point of view of both sides of the marketplace, and track it through to its actual outcome.Deliverable: A reliable outcome event, distinct from the request simply being sent.
- Step 3Build the supply-side intent signal from actions observed in the client's tools, not from their declarations.Deliverable: An intent score per listing or per seller, usable by the engine.
- Step 4Open an outbound recommendation channel on a limited segment, with a control group that stays on search alone.Deliverable: A readable comparison between incremental conversion and simple traffic displacement.
- Step 5Loop back on the actual outcome and remove already-filled items from the engine, so recommendation does not push what is no longer available.Deliverable: A retraining loop fed by outcomes, not by clicks.
- Step 6Frame automated pre-qualification: explicit criteria, human oversight on the final decision, disclosure to the person being assessed.Deliverable: A written, enforceable framework, a condition for deployment in Europe.
First step: Measure the current split of conversions between internal search and pushed channels. Without that starting point, you will not know whether recommendation adds conversion or displaces the conversion that already existed.
Sources
- S1 Transcript of Earnings Results Call for Q4 FY2025 - Recruit Holdings (transcription officielle) Primary archive pending
- S2 Indeed Helps 31 People Get Hired Every Minute, Powered by AI-Driven Matching and Hiring Innovation Primary archive pending
- S3 Indeed parent company touts record growth driven by AI - HR Dive Established press archive pending
- S4 Transcript of Earnings Results Call for Q1 FY2026 - Recruit Holdings (transcription officielle, preuve de vivacite) Primary archive pending
- S5 Indeed's AI-Powered Sourcing Assistant Helps Employers Hire Over 30% Faster Primary archive pending
- S6 Recruit Holdings Q4 FY2025 Earnings Call - transcription officielle en PDF (meme document que S1) Primary archive pending
An error, newer info, a source?
This page lives on its accuracy. If a figure has moved, if the deployment has changed, or if you have a higher-quality source, tell us. Every sourced correction is verified before publication.