In many companies the careers page is where the care that goes into every product page runs out. The job ad is online, an email address or a form sits underneath it, and the matter is considered settled. But the candidate reading it on the sofa at half past ten in the evening has questions. What is the salary range? How many days of remote work are realistic? Is there a shift pattern? Will a qualification earned abroad be recognised? And what became of the application sent three weeks ago? If those questions stay open, she drops off, not because the role is unattractive but because nobody was reachable. The labour market leaves little room for losses of that kind: in the fourth quarter of 2025 there were 1.26 million (IAB) job openings across Germany, and registered vacancies took an average of 180 days (Federal Employment Agency) to fill in January 2025. This article shows how an AI assistant on the public careers page answers candidate questions around the clock, guides people to the right role, captures short applications and writes interview slots into the HR calendar, and where its limits are drawn on purpose.
Key takeaways
- The careers page is often the last page without anyone to talk to. In the fourth quarter of 2025 there were 1.26 million (IAB) job openings across Germany, and registered vacancies took an average of 180 days (Federal Employment Agency) to fill in January 2025. Every unanswered question adds to that time.
- 79 percent (Nielsen Norman Group 1997) of users scan a page rather than read it. A job ad therefore rarely answers what candidates actually care about: salary range, share of remote work, shift pattern, recognition of foreign qualifications, application status.
- An assistant trained on job ads, the benefits page and cleared HR knowledge answers those questions around the clock, guides candidates to the right role with a few follow-up questions and captures a short application directly in the chat.
- The assistant writes interview slots into the HR team's calendar. Brevity is the lever here: an average retail checkout still contains around 11.3 form fields (Baymard Institute 2024), and better input design lifts completion rates at large shops by an average of around 35 percent (Baymard Institute 2016).
- What the assistant does not do stays clearly defined: no commitments, no contract details, no assessment of documents. On top of that come non-discriminatory wording under equal treatment law and fixed deletion periods for candidate data.
The careers page as a blind spot
Companies invest heavily in product pages, landing pages and contact channels for customers. The careers page runs alongside: a list of open roles, a PDF, an email address. Yet in many professions the competition for candidates is fiercer than the competition for customers. In the fourth quarter of 2025 there were 1.26 million (IAB) job openings across Germany, around 80 percent (IAB) of which needed filling immediately, meaning they were already vacant. In the first quarter of 2025 the vacancy rate stood at 2.6 percent (IAB). For comparison: in the fourth quarter of 2022 there were still around 2 million (IAB) openings. So the market is calmer than at its peak, but filling roles still takes time: registered vacancies stayed open for an average of 180 days (Federal Employment Agency) in January 2025. Every week a role stays open costs contribution margin, overtime and the team's patience.
The second reason for the blind spot is reading behaviour. 79 percent (Nielsen Norman Group 1997) of users scan each new page, only 16 percent (Nielsen Norman Group 1997) read word by word, and on an average page at most 28 percent (Nielsen Norman Group 2008) of the words are actually read. A job ad that carries all the answers somewhere in its body copy therefore answers very little. The question left in the candidate's head goes unasked, because there is no channel for it, least of all at half past ten in the evening. That is exactly the gap an assistant on the careers page closes. It should not be confused with the internal AI assistant for employees, which distributes knowledge inwards; this is about external candidates who have not applied yet.
In brief: what a recruiting assistant on the careers page does
The questions that actually get asked
Reading chat logs from careers pages turns up surprisingly few surprises and a great deal of repetition. Before applying, candidates want to know whether the effort is worth it. So they ask about the hard parameters, not about company culture. And they ask when they have time: after work, at the weekend, during a lunch break. The following topics come up regularly in careers page projects.
- Salary range: a band, a pay grade or at least a statement of what the grading depends on
- Working hours and remote work: how many days on site are expected and whether there are fixed core hours
- Shift pattern: which shifts exist, how long the rotation runs and how weekends are handled
- Recognition of foreign qualifications: which documents are needed and who supports the process
- Language level: which level of German or English the role genuinely requires
- Process and duration: how many conversations are planned, whether a trial day is involved and when feedback arrives
- Application status: what became of the application sent three weeks ago
These questions can be answered without a person stepping in, provided the answers sit cleanly and up to date in a maintained source. That is precisely why the knowledge base is the foundation: every answer comes from a cleared entry with a reference and a validity date, not from free-form phrasing. Which content belongs in it and how it should be cut is covered in building a knowledge base for AI chat. For the HR team that means something concrete: what the assistant is allowed to say is decided by an approval, not by chance.
Answers around the clock
Questions arise in the evening and at weekends. The assistant answers at the exact moment they come up, rather than on the next working day.
Guiding to the right role
Two or three follow-up questions on tasks, location and availability are typically enough to name the right posting.
Short application in the chat
Name, contact, preferred role and earliest start date are captured in conversation instead of in a long form full of mandatory fields.
Booking a slot directly
Open slots in the HR calendar appear in the chat, and the confirmation goes out to both sides straight away.
Application status
Once a person is reliably identified, the assistant names the processing status and the next step without revealing internal details.
Limits and data protection
Commitments, contract questions and the assessment of documents stay with the team, and candidate data is subject to fixed deletion periods.
From interest to the right role
Many careers pages list roles alphabetically or by department. For candidates who do not know the internal structure, that is a search task. An assistant reverses the direction: instead of asking the visitor to filter, it asks two or three short follow-up questions and then suggests one or two postings that genuinely fit. It is the same mechanism that works in product advice, only with a different goal: what stands at the end is not an order but an application that matches the role, which means less effort on both sides.
- Entry: the assistant asks about the field of work or the qualification, not about a job reference number
- Narrowing down: location, preferred hours and earliest start date already rule out most postings
- Suggestion: one or two matching roles, each with a sentence on the task and a link to the full ad
- Clarification: open questions on salary range, working hours or onboarding, answered from the knowledge base
- Short application: name, contact, preferred role, availability and optionally a document are captured in the conversation
- Handover: the case reaches the HR team in structured form, by email or directly in the applicant tracking system
The reason a short application in the chat reaches more candidates than the form next to it is friction. In retail an average checkout still contains around 11.3 form fields (Baymard Institute 2024); roughly 22 percent (Baymard Institute) of online shoppers have abandoned an order because the process felt too long or too complicated, and better input design lifts completion rates at large shops by an average of around 35 percent (Baymard Institute 2016). An application form with a CV upload, mandatory fields and an account to create is no different. How the same effect plays out in customer acquisition is described in qualifying leads via chat instead of a long form; the technology behind it is the lead assistant, which passes on the captured details in structured form. That the same route also carries complex enquiries is shown in the article on B2B quote requests in the chat.
The short application is the real lever
The interview slot goes straight into the calendar
The second big point of friction sits behind the application: scheduling. Three emails back and forth, plus a candidate who works during the day and cannot take calls freely, and days pass in which another company moves first. An assistant with calendar access offers genuinely open slots instead, blocks the chosen one straight away and sends confirmation and reminder. For the HR team the workflow barely changes: they maintain their calendar as before, the assistant reads the availability and writes the appointment in.
To work reliably, the assistant needs more than read access, it needs permission to carry out an action. How an assistant drives systems instead of merely producing text is explained in AI assistants controlling your tools; the scheduling mechanics in detail are covered in booking appointments via a chat assistant, and the matching service is the booking assistant. The effect on time to hire cannot be quantified across the board, as it depends on the sector, the role and the team. What holds up is the observation that the stretch between first interest and first conversation is where candidates are lost most often.
| Aspect | Careers page without an assistant | Careers page with an assistant |
|---|---|---|
| Evening questions | Stay open until the next working day | Are answered from cleared sources |
| Choosing a role | The candidate filters through the list alone | Two or three follow-up questions lead to the right role |
| Application | A long form with mandatory fields and an upload | A short application in conversation, documents later |
| Scheduling | Several emails spread over days | An open slot is booked bindingly straight away |
| Application status | A phone call or email is needed | An answer in the chat, once the person is reliably identified |
| Insights | Drop-offs stay invisible | Recurring questions reveal gaps in the job ad |
What the assistant deliberately does not answer
A recruiting assistant that is allowed too much does more damage than one that can do too little. A salary commitment made in a chat, a statement about an application's prospects or an opinion on submitted documents is not a service, it is a risk: it creates expectations the company is later expected to meet, and it pre-empts a decision that people make. Clear blocks therefore belong in the setup. The assistant states the framework and the process, but it does not negotiate, does not assess and promises nothing. A limit applies to the application status too: it names the processing status only once the person is reliably identified, and it confines itself to the next step rather than internal opinions.
An assistant that commits to a salary in a chat has not recruited faster, it has run a negotiation it had no mandate to run. That limit belongs in the concept, not in the clean-up afterwards.
Technically, this restraint rests on the assistant answering exclusively from cleared sources and passing on anything uncertain instead of phrasing whatever sounds plausible. How to secure that is described in preventing hallucinations with a clean knowledge base. When a conversation reaches the limit, an orderly handover to a human colleague is the decisive step. That also matches what users expect: 62 percent (Bitkom) want to reach a quickly available human contact when there is a problem, while a chatbot is preferred by only 36 percent (Bitkom). In recruiting this weighs double, because a person's professional future is at stake.
Equal treatment and non-discriminatory wording
An assistant in the hiring process speaks on behalf of the employer. Its wording is therefore held to the same standard as the job ad. Under section 1 (AGG) the German General Equal Treatment Act prohibits disadvantage on grounds of race or ethnic origin, gender, religion or belief, disability, age and sexual identity, and section 11 (AGG) expressly requires that a job advertisement must not breach the prohibition of disadvantage in section 7(1) (AGG). For the assistant that means three concrete rules: no follow-up questions about age, marital status, pregnancy, religion or origin; no wording that addresses one group while excluding another; and the same depth of information for every candidate.
In practice, equal treatment also means equal treatment of information. An assistant that gives one person the salary range and evades the question for another creates exactly the unequal treatment it is meant to avoid. Answers to sensitive framework questions are therefore stored fixed rather than left to the flow of the conversation. Personalisation has a different place here than in a logged-in area, where the assistant knows the orders and contracts of a known person; how that works there is shown in the article on personalised answers in the customer area. On a public careers page the opposite applies: everyone gets the same information, regardless of how the question was phrased.
Wording that has no place in an application chat
Candidate data, deletion periods and hosting
As soon as the assistant captures a name, contact details and a preferred role, it processes candidate data. The same obligations apply as in the rest of the hiring process. Processing rests on Article 6(1)(b) (GDPR) for steps prior to entering an employment relationship, the information on purpose, recipients and storage period follows from Article 13 (GDPR), and the storage limitation in Article 5(1)(e) (GDPR) requires that data is not kept longer than necessary. In practice this leads to the widespread period of around six months after the process ends: claims under equal treatment law must be asserted in writing within two months (AGG), the subsequent limitation period for filing suit is three months (ArbGG), plus a short safety margin. Anyone wanting to move candidates into a talent pool needs separate consent for that.
- Purpose limitation: chat content from the hiring context is processed and analysed separately from customer enquiries
- Data minimisation: the assistant asks only for what the next step requires, no identity document numbers, no health data
- Transparency: at the start of the conversation it is clear that an assistant is answering and where the details go
- Deletion periods: short applications and the associated chat logs expire automatically once the defined period has passed
- Roles and rights: only the HR team sees candidate data, and the analysis of conversations runs in aggregate
- Hosting: processing and storage take place in Germany, with a data processing agreement and documented deletion runs
How to implement this technically and contractually is explored in the article on GDPR-compliant AI assistants; the framework for operation and storage location is described on the data protection and hosting page. For employee representation it also matters that the analysis of chat logs serves to improve ads and answers, not to evaluate individual employees. An explicit assurance to that effect belongs in the rollout, as does a decision on who reviews the logs and at what interval.
Chat logs show where the job ad is unclear
The underrated benefit lies not in the chat itself but in the analysis afterwards. When twenty candidates ask about the shift pattern in a single week, it is missing from the ad. When the question about the salary range breaks off at the same point every time, the answer there is too vague. When a posting triggers many conversations but produces few short applications, something between expectation and description does not add up. Without an assistant these patterns are invisible, because an unanswered question leaves no record. With one, they become the to-do list for the next revision of the job texts.
The route there runs through conversation analytics, which summarises topics, drop-off points and gaps; how to derive concrete improvements from that is shown in analysing and improving chats. The occasion is not theoretical: 41 percent (Bitkom) of companies with 20 or more employees now use artificial intelligence, up from 17 percent (Bitkom) the year before, and a further 48 percent (Bitkom) are planning or discussing it. Dialogue with customers and candidates is one of the areas where an assistant becomes visible first. In the IW future panel, 84.5 percent (German Economic Institute Cologne) of the companies that use or plan to use AI name relief from routine work as the goal, while 37.0 percent (German Economic Institute Cologne) of the companies surveyed use AI at all. HR is among the areas with the highest share of recurring questions and is therefore an obvious starting point.
A few steps to an assistant on your careers page
Getting started is smaller than it sounds. It begins with a stocktake: which roles are open, which questions arrive today by email and phone, what is already in the ad and what is missing? Out of that comes the first version of the knowledge base, usually from job ads, the benefits page and a short collection of answers from the HR team. After that the limits and handover routes are defined, the assistant is embedded on the careers page with a short snippet and signed off with test cases. Two to three weeks later the first analyses show which ad needs sharpening. If you would like to work through this with your own postings, a no-obligation demo is the fastest route.
- Compile open roles, benefits and the steps of the process as cleared sources
- Collect the ten most common candidate questions and define one solid answer for each
- Define the block list: commitments, contract details and the assessment of documents stay with the team
- Connect the HR team's calendar and define which slots can be booked
- Put deletion periods for short applications and chat logs down in writing
- Run test cases with inadmissible questions and review the answers before go-live
Sources and Studies