SeekOut

Three systems SeekOut does not own receive its writes. The customer's mailbox: Recruit Core includes email integration and send on behalf, so SeekOut sends outreach under the recruiter's own identity and runs AI multi step campaigns through it.

The customer's ATS, in both directions: the Candidate Finder "searches public candidate data and ATS data", Rediscover Applicants exists to "reconnect with past candidates in your ATS", and Core includes Excel, PDF and ATS export, so records are read out and pushed back.

The security page describes the mechanism: "communication through ATS partner APIs is HTTPS encrypted using TLS 1.2+" with AES-256, a vendor describing a live authenticated integration rather than claiming one. And the open web: the Chrome extension sources and enriches profiles from any website.

The navigation names the partner estate as categories (ATS, CRM and HCM integrations), with custom integrations listed at the top pricing tier. The mechanism, direction and encryption are documented, but no individual ATS, CRM or HCM vendor is named.

Sourceseekout.com/pricing, /security and /capabilities/agentic-ai-recruitingread 2026-09-12

SeekOut runs sourcing, screening and engagement as one sequence. The positioning is "from job description to qualified shortlist in minutes", and the invitation is to "put AI agents to work across sourcing, screening, and outreach". Agents "independently research candidates, evaluate qualifications, craft outreach messages, and manage engagement sequences".

The stages are distinct, named and handed off in order: a rubric is built from the customer's criteria, candidates are found across public and ATS data, fit is analyzed against the rubric, outreach is written per candidate and sent, market data is researched alongside, and replies are tracked. The rubric ties the stages together: written once, it governs both the search and the scoring, a real dependency between steps.

Decisions are made mid flow rather than preconfigured, and SeekOut draws the contrast itself: "Traditional automation follows predefined rules: send email A on day 1, email B on day 3.

Agentic AI understands context and makes decisions: it adjusts search criteria based on market conditions, personalizes outreach based on a candidate's background, and prioritizes candidates based on multiple signals simultaneously."

The participants are distinct: the recruiter who sets the rubric and approves, six named agents with different jobs, the hiring team, and the candidate who replies into the tracked sequence.

An outside orchestrator can drive it. Through SeekOut MCP an external assistant runs the search and the recruiting work, and the top tier lists agentic AI workflows and job specific workflows and rubrics as named entitlements, so the flow is something a buyer configures rather than a fixed pipeline.

Sourceseekout.com/capabilities/agentic-ai-recruiting and /pricingread 2026-09-12

SeekOut grounds its agents in a standing index of more than one billion candidate profiles, searchable on every plan: unlimited search, 1B+ profiles, AI search and 30+ filters are listed in the entry tier. The figure is stated consistently across the navigation, the capability page and the pricing page. The corpus exists between runs and far exceeds any context window.

Retrieval is semantic, and the vendor says so: profiles are "read for context, not keywords", and the agents conduct "in-depth analysis across resumes, candidate profiles, and broader market data to uncover top talent that others might easily overlook". A Market Researcher agent queries talent market data as a separate source.

The customer's own data is in scope too. The Candidate Finder searches public candidate data and ATS data, and Rediscover Applicants searches talent already in the customer's ATS, so the index spans a public corpus SeekOut maintains and the buyer's own past applicants, queried together. Grounding can be checked per result: candidates are scored against a rubric the customer writes, with the reasoning shown. Talent pool insights and market data are part of the same corpus.

Sourceseekout.com/capabilities/agentic-ai-recruiting, /pricing and /capabilities/rediscover-applicants as named in navigationread 2026-09-12

Recruiters keep the hiring decision, but the outreach the agents send is not shown as gated. SeekOut says "recruiters maintain full control over hiring decisions", that "human recruiters ensure only the most qualified candidates move forward", and that agentic AI "partners with human recruiters". Its FAQ goes further: "recruiters approve or adjust at every critical step", with transparent explanations on every recommendation. An approval step at each stage is a real claim.

The consequential action is described as ungated. The outreach agent "creates personalized outreach based on candidate research and sends emails that convert"; the platform runs AI multi step campaigns with email integration and send on behalf, so SeekOut sends under the recruiter's identity; and agents work "around the clock to evaluate and engage hundreds of candidates at once". Nothing published says a message is previewed, held or approved before it reaches a candidate, and sending on a recruiter's behalf, unattended and at volume, is the action a gate would need to cover.

No named surface backs the approval claim either: no view for the approver, no pending approval queue, no escalation or timeout behavior, and no configurable autonomy level. The approval language sits in a marketing FAQ rather than a documented mechanism, and approving at every step and sending emails are both described on the same page.

Sourceseekout.com/capabilities/agentic-ai-recruiting and /pricingread 2026-09-12

Certification and access controls are both published, on a dedicated security page and a trust center at trust.seekout.com, each linked from the primary navigation. The SOC 2 claim reads "SeekOut is SOC 2 Type 2 certified as of June 8, 2023", restated in the FAQ as "SeekOut maintains SOC 2 Type II certification". Customer access control: "For customers who want unified access control, SeekOut supports SAML-based single sign-on provisioning systems", and the FAQ names the providers: "enterprise SSO through SAML 2.0, integrating with identity providers like Okta, Azure AD, and OneLogin."

The wider program is specific: annual third party penetration testing of the cloud environment, web applications and network configuration; monthly vulnerability scans; a public HackerOne program; a stated 24 hour breach notification commitment; bcrypt password hashing; AES-256 at rest and in transit with keys managed by Azure; an appointed Data Protection Officer; GDPR and CCPA compliance with data processing agreements and support for subject rights; and OFCCP record keeping and reporting for federal contractor customers. SeekOut commits that it "does not use any client data to train our AI models".

AI governance is claimed in weaker words: SeekOut states it is "compliant with ISO/IEC 42001:2023 and ISO/IEC 23894:2023", the AI management and AI risk standards. Compliant with is not certified, so those read as alignment claims. The certification list the page gives for Azure belongs to the cloud provider, and the internal Team Access Control policy, with role based access and quarterly reviews, governs SeekOut's own employees rather than the customer's users. The SOC 2 date is June 2023, though the FAQ says the certification is maintained.

Sourceseekout.com/security and its FAQread 2026-09-12

Every candidate score comes with its reasoning; the agents' own actions are harder to see. The capability page states that "every AI recommendation comes with transparent explanations", and that SeekOut "scores each one against the rubric you set, and shows the reasoning behind every result, so quality and compliance hold as volume rises". Because the customer writes the rubric, the explanation reads against a standard the buyer set rather than an opaque relevance score. Consistency is claimed on the same basis: "every candidate is evaluated using the same criteria."

A score explanation answers why this candidate; an audit trail answers what the agent did. No run log, action history, export, retention period or operator view of agent activity is published: which candidates the Candidate Finder discarded and on what score, when the outreach agent sent and under whose identity, or how the Engagement Tracker classified a reply.

The top tier lists benchmark reporting and dashboards. Reporting on hiring benchmarks measures the customer's recruiting funnel rather than the agents' behavior.

Sourceseekout.com/capabilities/agentic-ai-recruiting and /pricingread 2026-09-12

Learning is claimed, but memory is not documented. SeekOut's FAQ says its agentic AI "adapts search criteria, personalizes communication, and learns from results" and "adjusts search criteria based on market conditions". Nothing says what is learned is kept as state a customer could inspect; it is folded into how the next search behaves, which is a model rather than a memory. Refining a rubric across the steps of one search is context within that run, not persistence across sessions.

The Engagement Tracker "manages candidate interactions" and engagement sequences, holding who was contacted, when and what came back. That is a record of outreach, the customer's application data, not agent memory. The profile index is the knowledge corpus, also not memory.

Nothing describes a named memory layer, a retained recruiter preference or correction the agents reapply to later searches, a decision trace carried between runs, a stated scope or lifetime, or a customer control to inspect, edit, export or purge agent state apart from candidate and campaign records. The pricing page itemizes the platform down to 500 contact credits a month and 1,000 exports and names no memory component. When a recruiter drives SeekOut through MCP from Claude or Copilot, the conversation context lives in the calling assistant, not in SeekOut.

Sourceseekout.com/capabilities/agentic-ai-recruiting and /pricingread 2026-09-12

SeekOut runs in one cloud and one country, with no choice offered. It states it is "entirely hosted on Microsoft Azure servers", and its own FAQ answers the residency question: "SeekOut's infrastructure is hosted on enterprise-grade cloud platforms with SOC 2 certified data centers in the United States."

No selection surface exists. The pricing page itemizes four tiers down to contact credits, export caps, per seat mechanics and custom SLAs, and has no region, residency or hosting row at any level, not even in the top tier, where custom integrations and custom SLAs and compliance are the differentiators. No customer controlled environment, dedicated tenancy or self hosted option appears anywhere.

Geographic replication reads like multiple regions and is not. SeekOut states that backups "are geographically distributed" and that "most data is geographically replicated to enhance webapp performance and recovery during a disaster". That is resilience and latency engineering inside the vendor's own architecture, not a residency option the customer selects. A buyer in the EU with a data localization requirement has no documented option, which matters for a vendor that processes candidate personal data at scale under GDPR and appoints a DPO.

Sourceseekout.com/security and its FAQ, with /pricing; seekout.com/pricingread 2026-09-12

Named agents ship, but they are stages of one workflow rather than a catalog. The capability page's body names three agents (Rubric Creator, Candidate Finder and Outreach Expert), while its FAQ names six: Rubric Creator, Candidate Finder, Profile Analyst, Outreach Author, Market Researcher and Engagement Tracker. The outreach agent is Outreach Expert in one place and Outreach Author in the other, on the same page.

The six are the steps of one sourcing workflow: build the rubric, find candidates, analyze fit, write outreach, research the market, track replies. Remove the outreach agent and Candidate Finder is not a whole product a buyer adopts instead; it is the same platform with a gap. SeekOut describes them as a "suite of AI agents" and "our digital coworkers" that "partner with our recruiters", working "inside the platform your recruiters already use". None is installed, selected, removed or priced on its own.

What a buyer chooses among is on the pricing page: Recruit Core, Sourcing, Sourcing + Integration and Full Recruiting Funnel, with "mix and match any product modules". Agentic AI workflows appear as a line item on the top tier, not as a catalog entry. The agents do ship prebuilt and named, needing no build by the customer, and job specific workflows and rubrics are listed as included at the top tier.

Sourceseekout.com/capabilities/agentic-ai-recruiting and /pricingread 2026-09-12

Scheduled multi step sequences ship, but a person still starts the work and email is the only channel. Recruit Core includes AI multi step campaigns with email integration and send on behalf, and the agents "manage engagement sequences", a follow up structure that runs unattended.

SeekOut contrasts this with fixed timing ("send email A on day 1, email B on day 3"): it prioritizes candidates on multiple signals, so the cadence is decided rather than fixed. Agents "work around the clock to evaluate and engage hundreds of candidates at once", and the Engagement Tracker "manages candidate interactions", so a reply is an event the sequence responds to.

The initiating trigger is a person: a run begins from a job description or a recruiter request. Nothing documents a configurable event trigger, such as a new requisition, an ATS status change or a schedule the customer sets, starting the work without a recruiter asking.

Channel coverage is thin. Outreach is email; the entry tier lists email integration and nothing else, and no LinkedIn, InMail, SMS, voice or ticketing channel is offered. SeekOut MCP and Microsoft 365 Copilot put the product inside other assistants, which is a way in for recruiters rather than an outreach channel.

Sourceseekout.com/pricing and /capabilities/agentic-ai-recruitingread 2026-09-12

One provider is named and the customer chooses nothing. SeekOut's security page states: "SeekOut uses Azure AI services for our platform." The capability page describes the intelligence generically ("these large language models are trained on massive datasets") and names no model, version or alternative.

There is no model picker, admin setting, per agent assignment or bring your own key path, and the pricing page itemizes the platform down to contact credits, export caps and per seat mechanics across four tiers with no model row. Nothing says SeekOut routes across more than one model internally either.

Driving SeekOut from Claude, ChatGPT or Copilot through SeekOut MCP is not model choice. That is the customer's own assistant calling SeekOut; it does not change the model that runs SeekOut's rubric scoring, profile analysis or outreach writing. The Azure relationship does not make Microsoft a second model either: a cloud vendor is infrastructure.

Sourceseekout.com/security, /capabilities/agentic-ai-recruiting and /pricing; seekout.com/capabilities/agentic-ai-recruitingread 2026-09-12

An outside assistant can drive SeekOut through a shipped MCP server. SeekOut MCP sits alongside SeekOut Recruit and SeekOut Sam as one of three solutions in the primary navigation, described as "Use SeekOut inside Claude, ChatGPT, Copilot, and any AI assistant". Documentation has its own host at docs.seekout.com, promoted from the navigation as "setup guides and tools for SeekOut MCP", and the featured entry says the docs "cover setup, tools, and example prompts". A documented tool list is what separates a shipped MCP server from an announcement.

It reads, writes and triggers actions. A recruiter can "search and recruit" straight from the assistant they already use, running the recruiting work rather than just retrieving profiles, and the pricing page lists SeekOut MCP as part of what Recruit includes, so it is a shipped entitlement rather than a pilot. Any MCP compatible client is admitted, not a fixed list. A second inbound surface is separate from MCP: a site wide banner announces "SeekOut now in Microsoft 365 Copilot". The 14 day free trial gives full access to Recruit Core, so a prospect can reach MCP without a contract.

No separate public REST API reference, endpoint list, SDK or webhook surface is published.

Sourceseekout.com/solutions/mcp as named in navigation, docs.seekout.com as published, and /pricingread 2026-09-12

Nothing tests the agents themselves. SeekOut scores every candidate against a rubric the customer writes and shows the reasoning, and says "every candidate is evaluated using the same criteria". That is the agent's output about candidates, not a measure of the agent; consistent criteria are a property of the rubric, not a test result.

The headline figures measure the customer's recruiting operation: "5X more candidates evaluated vs traditional methods" and "50% faster time-to-fill", and the top pricing tier includes benchmark reporting and dashboards. Time to fill and hiring benchmarks are funnel metrics, not evaluations of agent behavior.

No test environment, sandbox, holdout, A/B comparison, evaluator, prompt or rubric versioning, or scored verdict about agent behavior is published, and the four tier pricing page itemizes the platform down to monthly contact credits without naming one. Nothing says a changed rubric, search or message template is tried against a known set before it reaches candidates.

Sourceseekout.com/capabilities/agentic-ai-recruiting and /pricingread 2026-09-12

A recruiter does the browsing here, not the agents. SeekOut's Chrome extension is described as "source and enrich profiles from any website": a recruiter browses to a page, and the extension captures and enriches what is on it. That is an ingestion path a person operates, not the product navigating anything.

The agents work against a maintained profile index and authenticated ATS and email APIs, not against rendered pages. Where the extension does read a page, a breakage is a capture failure the recruiter sees on a page they chose to open, not an unattended agent losing its way mid run.

Nothing else claims the capability: no hosted or local browser, desktop session, remote or local computer control, RPA or recorder is published, and the pricing page itemizes the platform down to export caps and contact credits without naming one. Outreach reaches candidates through email integration rather than by operating a webmail interface, assistants reach SeekOut through MCP rather than by driving a browser, and no LinkedIn send path appears anywhere on the site.

Sourceseekout.com/chrome-extension as named in navigation, with /capabilities/agentic-ai-recruiting and /pricingread 2026-09-12