Candidates Are Researching You With AI — AI Perception Management for HR and Employer Branding

Candidates are using AI to research employers. In a Japanese survey by Mynavi, 84.9% of students graduating in 2027 had used AI in their job search. Employers are not unaware of this either: in another Mynavi survey, 30.6% of companies said they had a sense that students were using generative AI, up from the previous year.

The problem lies past that point. Recognizing that students use AI and measuring how your own company is described on AI are two different things. The first can be learned from industry surveys. The second appears in none of them, because the only way to obtain it is to measure it yourself. The subject of this article is not an awareness gap but this measurement gap.

Candidates do not swallow AI answers whole. In the same survey, among the 1,184 students with experience using AI, 2.9% said they use what AI suggests almost as-is as their own decision, while the most common answer, at 68.7%, was that they treat it as one input among several. But an input is still an entry point: the frame AI presents becomes where consideration starts.

That is why a company needs a continuing view of what is being said about it on AI. This article is a practical guide for building one.

A note on scope. The evidence in this article is drawn almost entirely from Japanese survey data published in 2025 and 2026, and it describes the Japanese hiring market. Readers outside Japan should treat the figures as indicative rather than directly transferable.

This article is written for HR and recruiting professionals, employer branding practitioners, and CHROs and heads of HR at operating companies. It is not a guide to using AI for students or job seekers. It starts from the premise that candidates are researching your company with AI, and asks what the company needs to know and to put in place.

1.1 AI use in job hunting has reached 84.9%

According to a Japanese survey conducted by Mynavi from April 25 to 30, 2026 — the "April 2026 Career Intentions Survey of University Students Graduating in 2027 (On AI Use Among Job Seekers)," with 1,258 valid responses from university and graduate students nationwide expected to graduate in March 2027, conducted online via web direct mail to Mynavi 2027 members — 84.9% of students had used AI in their job search.

Source: Mynavi, "April 2026 Career Intentions Survey of University Students Graduating in 2027 (On AI Use Among Job Seekers)" (published May 26, 2026)

From 18.4% to 84.9% in three years. This is no longer a story about a subset of students.

1.2 Reading the breakdown correctly

That 84.9% covers AI use across the job search as a whole, and the uses vary widely. The table below shows how the 1,076 respondents who said they had used AI in their job search answered a question about specific uses (multiple response). Note that the denominator is not the full 1,258 respondents.

Source: same as above (n=1,076; prior year n=928; multiple response in both)

The single most common use is polishing application essays, so writing support remains the center of gravity. But the three rows that matter for HR are the bottom three. Among students who had used AI, the share using it for industry research rose from 35.5% to 52.2% in one year, and the share using it to search for companies and jobs rose from 16.2% to 26.3% — meaning that more than one in four AI users now uses AI to find the companies and roles they will apply to.

That said, "industry research" does not necessarily mean looking up a specific company by name; it also covers researching an industry's overall direction.

The same survey also asked about use at decision points: 24.5% for considering which companies and roles to apply to, 9.4% for deciding which companies to interview with, and 8.2% for deciding which selection processes to withdraw from (all of the 1,076, multiple response).

1.3 Employer awareness is rising too

What about the employer side?

According to Mynavi's "2025 New Graduate Recruiting and Job Market Review (Class of 2026)," published on September 26, 2025, 30.6% of companies said they had a sense that students were using generative AI in their job search — above three in ten. That is up from 12.3% at the time of the survey covering the class of 2025.

Employer awareness is catching up. The premise that companies have not noticed students using AI does not hold, at least as of the class of 2026.

One caution. The 84.9% on the student side (class of 2027, surveyed in 2026) and the 30.6% on the employer side (class of 2026, surveyed in 2025) come from different survey years, different questions, and different populations. They cannot simply be subtracted to yield a "54.3-point gap." Nor could we confirm a published employer figure for the class of 2027 that would permit a continuous comparison using the same question.

1.4 Not an awareness gap, but a measurement gap

What employers have is awareness of a phenomenon: a sense that students appear to be using AI. What recruiting practice requires is something more specific — how your own company is described on AI. The first can be learned from industry surveys. The second is in none of them, because only the company itself can measure it.

Recognizing that candidates use AI and measuring how AI describes your company are separate problems. Noticing the first is a starting point, not a destination.

2.1 Very few take it at face value

You will encounter the claim that students take AI answers at face value. The data does not support it. In the Mynavi survey cited above, 1,184 students with experience using AI were asked how much AI output influences their decisions and actions. The results:

Source: same as above (n=1,184)

Only 2.9% use it almost as-is. The same survey asked respondents to rate on a 0–100% scale how far they found AI's answers useful or relatable; the average was 51.2%. Neither overconfidence nor distrust — a middling assessment.

2.2 Among those who consulted AI, more report an effect

A different question in the same survey shows another side.

Among the 576 respondents who had consulted AI about job-search concerns, when asked how much AI's answers had affected their judgment or actions, 13.8% said it affected them a great deal and 64.7% said it affected them somewhat — 78.5% in total reporting some effect.

However, this 78.5% and the figures in the previous section have different populations and different question formats, so they cannot be compared directly. The previous section asked all 1,184 AI users about influence in general; this one asked the 576 who actually consulted AI to look back on the effect of that consultation. It is natural for a group engaged enough to consult AI to report more influence, and this does not demonstrate a psychological paradox.

Read together without straining, the two are consistent with a single picture: few candidates adopt AI's suggestion wholesale, while most of those who actually consulted it recognize some effect. Not delegation, but an input.

2.3 First impressions, and the limits of that reading

In a Japanese survey conducted by Nyle Inc. from March 25 to 30, 2026, covering 419 students in the classes of 2026, 2027, and 2028 who had done or were doing a job search, respondents were asked how much a company summary generated by AI (features, strengths, weaknesses) affects their first impression of that company. The most common answer was "for reference only" at 65.6%, but 19.4% said it has a large effect — roughly one in five. The same survey found that around nine in ten students verify facts on the official corporate or recruiting site after researching a company with generative AI (32.3% always verify, plus 56.1% usually verify).

Nyle Inc. provides LLMO consulting, so this is a survey by an interested party (source: research by Nyle's SEO Consultation Room / https://www.seohacks.net/column/30459/).

It is tempting to conclude from this that a first impression survives verification, but that cannot be established from the surveys cited here. All of them are point-in-time attitude surveys rather than longitudinal studies. Whether an initial impression persisted independently of the verification, or whether verification confirmed a concern that had already been raised, cannot be distinguished.

What can be said in practice stops here: AI answers are entering the candidate's consideration process, and the company cannot see their contents.

The data in this chapter covers mid-career job changers, not new graduates. Do not read it as one continuous trend alongside the new-graduate data in the preceding chapters.

3.1 The nature of the survey

The respondents to the Japanese survey conducted by LANY Inc. from April 1 to 3, 2026 — the "Survey on Silent Withdrawal Caused by AI in the Recruiting Process," with 111 valid responses — were people aged 25 to 45 who had conducted a job search within the past year and who, in the course of it, had used a conversational generative AI service such as ChatGPT, Gemini, or Perplexity to research information about a specific company.

Because the population is limited to people who researched companies with AI, and the sample size is small at 111, the figures below are indicative values. They cannot be extrapolated to job seekers as a whole. LANY also provides LLMO support in the recruiting domain, so this too is a survey by an interested party.

3.2 What the negative information consists of

92.8% (n=111) said they had encountered information that left a negative impression when they asked AI about a company. The breakdown, from multiple response among the 103 who answered yes, was as follows.

Source: LANY Inc., "Survey on Silent Withdrawal Caused by AI in the Recruiting Process" (https://www.lany.co.jp/lany-llmo-lab/job-candidate-ghosting)

The top items are working hours, culture, pay, and turnover. Each of them is a topic a recruiting site's FAQ can get ahead of, and each is an area where companies tend to avoid publishing quantitative information.

The 17.5% for information that was abstract and hollow deserves attention as well. Nothing negative had been written; learning nothing was itself what registered as negative.

3.3 Those who withdrew most often cited comparison against competitors

In the same survey, 87.4% (n=111) said they had abandoned an application or withdrawn from a selection process because of information from AI. From multiple response among those 97 respondents, the characteristics of the AI information that triggered the withdrawal were as follows.

Source: same as above

The item chosen most often was being rated unfavorably in comparison with competitors. More than what is written about the company on its own, respondents cited where it was placed once set beside its competitors.

This suggests a difference from search engines. In conventional search, candidates looked at their own and competitors' sites separately and made the comparison in their own heads. Generative AI presents that comparison in a single answer. The frame of the comparison, and its conclusion, are assembled on the AI side.

3.4 The most frequently cited inaccuracy was outdated information

In the same survey, 91.5% (n=94) had experienced company information presented by AI that they felt was untrue or inaccurate. The results of multiple response among those 86 respondents were as follows.

Source: same as above

The item chosen most often was outdated information that had already changed. Because this is multiple response, it does not mean that 58.1% of all misinformation was outdated information, nor are outdated statements and baseless statements classified exclusively of one another. Even so, the fact that staleness was the most frequently cited problem in this limited survey is instructive in practice.

There is a lag before program changes and business reorganizations are reflected in the web information ecosystem. That characteristic is good news for tractability. This is not a matter of resigning oneself to the idea that AI errors are unavoidable: part of the problem reduces to the familiar operational task of keeping information current. Updating your own site does not resolve it immediately, however. The next chapter explains why.

For the definition, typology, and structural causes of misinformation generated by AI, see AI misinformation and brand misattribution, which covers the subject systematically.

4.1 The composition of cited sources

If you put the information written about your company in order, will AI's answers change? What has to be understood first is the composition of the sources AI draws on.

PerceptionX, which provides an employer reputation service, collected more than 100,000 AI answers to questions about employers between September 2025 and May 2026, and analyzed the 78,311 of them that contained at least one citation. According to the company, Glassdoor appeared in more than a third of the AI answers about major employers.

The analyzed set was collected from features that display their citation sources. Note that it is not a share breakdown of AI answers in general, which would include ordinary chat responses that display no citations.

In an article contributed to HR Executive by Karim Al Ansari, the company's co-founder, the sources AI draws on are grouped into four categories. Across the enterprise employers the company analyzed, owned media (careers pages, corporate blogs, official LinkedIn) accounted for 25% of AI citations, influenceable third-party sites (Glassdoor, Indeed, Comparably) for more than 40%, and community-derived sources (Reddit, Quora, Blind) for around 20%. The article also notes that outlets such as Business Insider, Fortune, and Forbes, along with best-workplace rankings, appear higher than the community platforms many HR teams watch closely.

However, the number of companies underlying this breakdown and the calculation method have not been published. PerceptionX provides a service in this domain, so this is an analysis by an interested party. The 25% and 40% figures cannot be treated as established fact; they should be read as an auxiliary line indicating a tendency. The company's analysis also covers mainly English-speaking enterprise employers, and it has not been confirmed whether the same composition applies to questions asked in Japanese.

For reference, in the LANY survey cited above, the most common action taken toward a company suggested by AI was checking its reputation on review sites or social media at 58.1%, ahead of looking at the official corporate or recruiting site at 36.2%. This does not verify the composition of AI citation sources, but it does provide supporting evidence that review sites and social media matter in candidates' verification behavior.

4.2 The implication — owned media does not decide the outcome

Putting your recruiting site in order is necessary, but it does not by itself determine AI's answers. AI assembles its answers by also referring to places the company does not control: review sites, communities, news coverage, ranking articles.

Many companies have concentrated budget and effort on the recruiting site and owned media, on the assumption that candidates eventually arrive there. But when a candidate puts the first question to AI, a summary built from third-party information has already been presented before they reach the owned media.

That brand perception on AI is not determined by a company's own output alone is the core problem addressed by AI Perception Management.

4.3 The discovery-channel side

AI does not bring only risk.

In the LANY survey cited above, 94.6% (n=111) had been suggested a company they had not previously known about as a result of consulting AI, and 31.4% of them actually applied. In the Nyle survey, around seven in ten job-seeking students said generative AI had served as an occasion for learning about a company (21.0% often, plus 49.8% sometimes).

Whether your company comes up in response to a question like "which companies in this industry are strong at X" can determine whether a first point of contact with a candidate exists at all. For companies with less name recognition, this may be an opportunity rather than a threat.

Caution is warranted, though. Both are small surveys by interested parties, and within the public materials we reviewed, we found no independent primary case study quantifying how much AI-driven awareness contributed to application volume or hiring outcomes.

5.1 What makes the recruiting context difficult

First, much of the information is hard to adjudicate. Statements such as "the hours are long" or "the culture would not suit you" are evaluations rather than factual errors. Parts that can be rebutted with numbers — average overtime hours, paid leave take-up, turnover — sit mixed together with parts that remain matters of judgment.

Second, the source is often an anonymous personal account. A review left by a former employee was that person's experience, and even if the company considers it skewed, it can rarely be declared false.

Third, candidates do not ask the company. In the LANY survey cited above, only 20.6% of the 97 respondents who withdrew had checked directly with a recruiter or interviewer before deciding. Most researched on their own, decided, and left without telling the company anything. No feedback comes back to the employer — that is the core of what LANY calls silent withdrawal.

The same survey found that among the 97 who withdrew, 3.1% used AI's information as a basis for their decision without verifying it, meaning 96.9% carried out some form of verification. Note, though, that this 96.9% takes the 97 who withdrew as its denominator, which is a different denominator from the 87.4% of all 111 respondents who reported having withdrawn. The two cannot be placed side by side and read causally.

5.2 How to think about a response

How to deal with misinformation generated by AI is a large subject in its own right. See the related articles below.

• For the definition and typology of misinformation, and an overview of why it arises structurally: AI misinformation and brand misattribution
• For the legal position, the legal avenues open to companies, and the debate over where responsibility lies: AI misinformation and legal liability
• For the practical work of filing corrections with individual platforms and getting information put right: Correcting and removing AI misinformation

Two points matter most for a recruiter.

An outside company generally cannot directly edit the answers of a third party's AI service at will. What it can do is combine the feedback and correction procedures each AI service offers with work on the sources AI refers to.

And correcting outdated information is the easiest place to start. In the limited survey discussed in the previous chapter, outdated information that had already changed was the most frequently cited type of inaccuracy. Cases where a program change or business reorganization has not been reflected can be improved by republishing accurate primary information. That is easier to take on than dealing with false and malicious statements.

Peer-reviewed research on recruiting and AI is accumulating. However, within the literature and public materials we reviewed, we found no peer-reviewed study that directly addresses this article's subject — candidates researching companies with generative AI. What follows is knowledge from adjacent areas, and it should be consulted with an accurate understanding of its scope.

6.1 The use of AI in the hiring process can affect organizational attractiveness

A study by Tursunbayeva, Fernandez, Gallardo-Gallardo, and Moschera was published in the European Management Journal in March 2025 (DOI: 10.1016/j.emj.2025.03.002). Using vignette experiments in two EU member states, it examined how the combination of AI in recruitment with professional and personal digital data affects candidates' perceptions of organizational attractiveness and their intention to apply.

The main findings are that the use of AI and digital data in recruiting can affect organizational attractiveness and application decisions, and that the effect differs by sub-dimension of organizational attractiveness and by the candidate's academic background. The study reports that engineering candidates showed lower perceived organizational attractiveness when told their digital data might be used in recruiting, while business candidates' intentions did not change. These are results under specific experimental conditions and sub-scales, and they do not support a generalization such as "engineering talent dislikes AI in hiring."

What this study addresses is companies using AI in recruiting, not candidates using AI to research companies. The direction is the reverse, and the two must not be conflated.

Even so, it is suggestive that the character of information within the hiring process can move perceptions of organizational attractiveness, and that the reaction is not uniform across candidate attributes.

6.2 The paradox of saying "we use AI"

A study by Oliver Schilke (University of Arizona) and Martin Reimann was published in volume 188 of Organizational Behavior and Human Decision Processes (article number 104405, DOI: 10.1016/j.obhdp.2025.104405). Across 13 experiments, it examined how disclosing the use of generative AI affects trust in the party making the disclosure.

The conclusion is consistent. Individuals and organizations that disclosed AI use were trusted less than those that did not. The authors call this the transparency dilemma and explain it as a decline in perceived legitimacy. The contexts tested include a job application setting.

This finding is a caution against the naive assumption that presenting yourself as a company enthusiastic about AI will simply communicate sophistication.

There are limits, though. This is not a study that examined employer branding claims about AI use directly. Nor is it a study that concluded disclosure should be avoided; the authors themselves argue for the need to lower the cost of transparency. The experiments involved mainly Western participants, and whether Japanese candidates would react the same way has not been tested.

And crucially, disclosure obligations arising from law, internal rules, or contract are a separate question from the effect of disclosure on trust. Where disclosure is required, it cannot be avoided on the grounds of its effect on trust. Nor does this study show that trust erosion can be avoided by wording things carefully.

One note: while writing, a study to the effect that stating transparency about AI use in job advertisements raises trust in the employer came up as a candidate source. It was not used, because its existence could not be confirmed against primary information.

7.1 Start by finding out — check under controlled measurement conditions

The first step is not a tactic but a check. Put the questions a candidate is likely to ask into several AI services.

• "What is it like to work at [company name]?"
• "Tell me about overtime and working styles at [company name]."
• "How does the culture at [company name] differ from [competitor]?"
• "Which companies in the [industry] industry are good places to work?" (does your company come up?)
• "Are there things to watch out for when applying to [company name]?"

The fourth and fifth matter most. The former tests whether your company can be discovered; the latter tests for the unfavorable comparison against competitors that respondents who withdrew cited most often.

And if you are going to check, record the measurement conditions. "We tried it a few times" gives you no way to isolate the cause when something changes, and no way to share the result internally. At a minimum, capture the following.

Only once these are recorded does a judgment such as "this is worse than last month" or "we are behind our competitor" have any basis.

7.2 Put your primary information in order

Prioritize the areas where AI most often gets things wrong. In the limited survey discussed above, the most frequently cited problem was outdated information that had already changed, followed by programs or benefits that do not exist. Check whether descriptions of programs, benefits, and business activities are current, and whether descriptions of discontinued programs still linger on the web.

Answer the topics candidates worry about with specific numbers. The negative information cited most often concerned hours and work-life balance, culture, pay, and turnover. There is value in publishing actual figures within the range you can disclose — average overtime hours, paid leave take-up, turnover, parental leave take-up — with the basis and the measurement period stated. Abstract phrasing such as "an open culture" can fall into the same category as the survey response about information that was abstract and conveyed nothing about the company.

State the update date, and get ahead of concerns with an FAQ. Showing what point in time the information reflects is useful to human readers and to AI. Organizing the topics candidates worry about in a question-and-answer format is also useful — but, as explained below, do not expect that attaching structured data to an FAQ will give you an advantage in how search engines display it.

7.3 Do not ignore third-party venues, and do not manipulate them

Since sources other than your own site account for a large share, you cannot ignore them. But there is a clear line here.

What you may do

• Publish accurate facts as official information on your company page at review sites (many services offer an official-information feature for employers)
• File through each service's proper, terms-compliant procedure regarding posts that contain factual errors
• Refrain from obstructing truthful, voluntary posts by employees
• Increase the amount of accurate information in third-party venues through interviews and contributed articles in outside media
• Cooperate with industry association surveys and official statistics so that objective data exists in a referenceable form

What you must not do

• Instruct employees to post favorable reviews, or ask them to do so in connection with compensation or performance evaluation
• Ask third parties to post reviews on your behalf
• Attempt to have negative reviews removed across the board regardless of whether their content is true

On employee posts, the reasoning in the Consumer Affairs Agency's Q&A on stealth marketing is a useful reference under Japanese rules. That Q&A explains that an employee posting impressions of their employer's products on their own social media is not, by that fact alone, immediately treated as a representation by the business, while noting that depending on the actual circumstances — the employee's position, standing, authority, duties, and the purpose of the representation — the business may be found to have been involved in determining the content, in which case it does constitute a representation by the business. In that case it must be clearly labeled as advertising or the like, so that its character as a representation by the business is unmistakable.

So this is not a matter of simply adding a note about one's employer. The practical points are not tying posts to company instruction, evaluation, or compensation; clearly disclosing the relationship where the post amounts to advertising or promotion; and taking care with confidentiality obligations and personal information.

Note that this notification sits within Japan's Act against Unjustifiable Premiums and Misleading Representations, which covers representations concerning goods and services offered to general consumers. Whether reviews in the recruiting domain fall within it depends on the content and purpose of the representation, so no blanket answer is possible. Even setting the legal question aside, however, making a representation in which the business was involved look like a third party's spontaneous post risks serious damage to the employer brand if it comes to light, and it may also violate the terms of the services concerned.

The legitimate approach is to keep publishing accurate primary information, in a form that is easy to reference, in sufficient volume. It looks like the long way around, but there is no other sustainable method.

7.4 Technical misconceptions to avoid

JobPosting structured data should be implemented for the job search experience, but it is not a dedicated measure that guarantees inclusion in AI Overviews or AI Mode. Google specifies JobPosting structured data as a requirement relating to the job search experience. On inclusion in AI features, the official Google Search Central documentation states that there are no additional requirements and no special optimization is needed, and that there is no need to create new machine-readable files, AI-oriented text files, or markup, and no special schema.org structured data to add. JobPosting structured data should be implemented properly for the job search experience. This point will be covered in detail in a separate article.

Installing llms.txt cannot be recommended as a measure aimed at Google Search. As above, Google has stated that creating new machine-readable files is unnecessary.

FAQPage rich results have not been displayed in Google Search since May 7, 2026. FAQ content itself is useful to candidates and can serve as material AI refers to, so there is a reason to create it — but you should not invest effort on the premise that adding structured data will make you stand out in search results.

Meanwhile, the basics Google officially recommends apply to recruiting sites without modification. Crawling should not be blocked by robots.txt or by your CDN or hosting layer; important information should be retrievable as text; structured data should match the visible text. Unglamorous, but these are the preconditions to check first. If your recruiting site cannot be crawled, then at least the measures intended to make that page referenceable in Google Search and Google's AI features will not work (information about your company on third-party sites is not affected by this).

One further point: in job postings that pass through recruiting agencies, staffing firms, or an applicant tracking system, the question of who the employer is creates issues specific to perception on AI. That will also be handled in a separate article.

7.5 Coordination across departments

How your company appears on AI is not something HR can manage alone. The sources consulted include press releases managed by corporate communications, disclosure materials from investor relations, and the corporate site managed by marketing. Conversely, information from the recruiting site is sometimes used in AI answers in non-recruiting contexts.

If HR tidies up the recruiting site on its own while other departments' output contradicts it, AI will present that contradiction to candidates along with everything else. At a minimum, we recommend a structure for sharing information about perception with the communications function.

8.1 Why continuity is necessary

First, because answers vary. The answer to the same question can change from run to run. A single check yields a single sample.

Second, because the sources change. New reviews, news coverage, discussion in communities — information rises and falls in places the company has no involvement in, and answers follow.

Third, because the AI services themselves are updated. An answer that raised no problem yesterday can change.

8.2 Do not treat "AI" as monolithic

A caution is needed here. "AI" is not a single thing.

Ordinary chat, chat with web search, Google AI Overviews, and Google AI Mode differ in how they access current information, whether they display citation sources, and how personalization and conversation history take effect. The points in the previous section — that answers change every time, and that history has an effect — can also vary in degree by service and by setting.

So you cannot roll everything up into a single statement that "this is how our company appears on AI." Which service, which feature, under which conditions — that is why the record of measurement conditions in section 7.1 becomes necessary.

8.3 What to watch continuously

"Do we show up on AI?" is not enough. You need to see what is said once you do show up, and how you are positioned alongside other companies.

8.4 The AI Perception Management framework

Grasping and managing your company's perception in the AI space on a continuing basis — across multiple AI services, including content and sentiment, under consistent conditions — is what Vaipm calls AI Perception Management (AIPM). Recruiting is one application domain of it.

The same framework applies to corporate communications and reputation, investor relations, and sales and marketing, because the sources consulted are shared across departments.

For the definition of the concept itself and its relationship to adjacent terms such as AIO, GEO, and LLMO, see What is AI Perception Management (AIPM)?, along with What is AIO (AI search optimization)? and What is LLMO (large language model optimization)?.

The surveys cited here share a set of limitations. Take note of them before citing any of this internally.

None of them track actual usage logs or behavior in an applicant tracking system; they rest primarily on respondents' self-reported answers. Figures such as "I stopped applying because of AI" or "it affected my first impression" indicate not a measured causal effect but the fact that the respondent perceived it that way. What actually moved the decision cannot be identified.

In addition, LANY, Nyle, and PerceptionX are all surveys by companies that provide services in this domain, and the sample sizes are limited, ranging from 111 to 419. PerceptionX's job seeker survey (306 respondents across seven countries) is restricted to job seekers who were using AI chatbots.

The figures in this article are therefore best treated as a starting point for grasping tendencies and identifying the questions to check for yourself. They are not a basis for asserting what the state of an industry is. Which is precisely why you need to measure your own company yourself.

Q1. Are students really researching companies with AI? Aren't they just using it to write application essays?

Both are happening. In Mynavi's Japanese survey of the class of 2027 (fieldwork April 2026, overall n=1,258), among the 1,076 respondents who said they had used AI in their job search, the most common use was polishing application essays at 71.8% — but researching companies is at a substantial scale too: industry research 52.2%, role and job research 35.8%, and searching for companies and jobs 26.3% (multiple response). Industry research rose from 35.5% the previous year, and searching for companies and jobs from 16.2%. Note that "industry research" is not synonymous with looking up a specific company by name.

Q2. Do students believe what AI tells them?

Few take it at face value. In the same survey, among the 1,184 students with experience using AI, 2.9% said they use AI's suggestion almost as-is as their own decision, while the most common answer, at 68.7%, was treating it as one input among several. Separately, among the 576 respondents who had consulted AI, 78.5% said it affected their judgment or actions — but the populations and question formats differ, so the two cannot be compared directly. It is natural for a group engaged enough to consult AI to report more influence.

Q3. What negative information does AI actually produce about our company?

In a Japanese survey by LANY Inc. covering mid-career job changers (fieldwork April 2026; 111 respondents aged 25 to 45 who had researched companies with AI), multiple response among the 103 who said they had encountered negative information put long hours or poor work-life balance at 63.1%, culture and workplace relationships at 35.9%, low pay at 30.1%, and high turnover at 27.2%. This is a small, limited survey and the figures are indicative. Note also that the respondents were not new graduates.

Q4. What should we do if incorrect information appears on AI?

An outside company generally cannot directly edit the answers of a third party's AI service at will. What you can do is combine the feedback and correction procedures each AI service offers with work on the sources AI refers to. In the LANY survey cited above, outdated information that had already changed was the most frequently cited type of inaccuracy, so correcting descriptions that do not reflect program changes or business reorganizations is an accessible starting point. For procedures and legal questions, see Correcting and removing AI misinformation and AI misinformation and legal liability.

Q5. If we improve our recruiting site, will AI's answers change?

It is necessary, but on its own it is likely to be insufficient. In PerceptionX's analysis, owned media accounted for around 25% of AI citations across the enterprise employers the company analyzed, with third-party sources such as review sites taking a much larger share. Note, however, that the underlying number of companies and the calculation method have not been published, that this is an analysis by an interested party, and that it covers mainly employers in English-speaking markets. Both are required: putting your own site in order, and creating a state in which accurate information exists in third-party venues.

Q6. Our ratings on review sites are low. How should we handle that?

Use the official information features that these services provide for employers, and file through their proper, terms-compliant procedures where there are factual errors. What you should avoid is instructing employees to post favorable reviews, asking them to do so in connection with compensation or performance evaluation, or having third parties post on your behalf. Under Japanese rules, the Consumer Affairs Agency's Q&A on stealth marketing indicates that even an employee's post may constitute a representation by the business depending on the actual circumstances — the person's position, standing, duties, and the purpose of the representation. Such conduct may also violate the terms of the services concerned.

Q7. Do JobPosting structured data or FAQ pages give us an advantage in AI search?

JobPosting structured data should be implemented properly for the job search experience, but it is not a dedicated measure that guarantees inclusion in AI Overviews or AI Mode. On inclusion in AI features, Google states that there are no additional requirements, that no special optimization is needed, and that there is no special schema.org structured data to add. As for FAQs, the content itself is useful for getting ahead of candidates' concerns, but FAQPage rich results have not been displayed in Google Search since May 7, 2026.

Q8. Does the approach differ between mid-career and new-graduate hiring?

The timing at which information is sought differs. In the LANY survey (mid-career job changers), the most common point at which people researched a company with AI was after receiving a scout email or an offer, at 53.2%, followed by while hesitating over whether to apply at 40.5%, and before an interview at 38.7% (111 respondents, multiple response). In mid-career hiring, the moment just after a scout message is sent is a critical juncture. We have not been able to confirm equivalent time-series data for new graduates, but AI is used across a wide span, from searching for companies through preparing for selection. The underlying principle — put accurate primary information in order — is the same.

Q9. What should we watch out for when checking AI's answers ourselves?

Record the conditions: the AI service and model, whether web search was enabled, language and region, date and time, login state and conversation history, the prompt, and the number of repetitions (see the table in section 7.1). Without a record, you cannot isolate the cause of a change. Also, ordinary chat, chat with web search, AI Overviews, and AI Mode differ in how they access current information and how personalization takes effect, so you cannot roll everything up into a single statement about how your company appears on AI.

Q10. Is it effective employer branding to say that we actively use AI?

This deserves careful thought. In research by Schilke and Reimann published in Organizational Behavior and Human Decision Processes (2025, 13 experiments), individuals and organizations that disclosed their use of generative AI were consistently observed to be trusted less than those that did not. That said, this is not a study that examined employer branding claims about AI use directly, and its participants were mainly Western. And where law or internal rules require disclosure, it cannot be avoided on the grounds of its effect on trust.

1. Students moved, and so did employers. AI use in job hunting reached 84.9%, and the share of employers who sense it rose to 30.6% (the survey years and populations differ, so the two cannot be compared directly).
2. But something is still not being measured. Knowing that students use AI and measuring how your company is described on AI are separate problems. The second appears in no industry survey.
3. Candidates do not swallow it whole, but they do use it as an input. Only 2.9% use AI's suggestion almost as-is. At the same time, the frame AI presents becomes where consideration starts.
4. Those who withdrew most often cited comparison against competitors. In a limited survey of mid-career job changers (97 respondents, multiple response), 69.1% cited being rated unfavorably in comparison with competitors.
5. The most frequently cited error was outdated information. In the same survey (86 respondents, multiple response), 58.1%. Part of the problem reduces to the familiar operational task of keeping information current.
6. Your own site does not decide the outcome. AI also consults review sites, communities, and news coverage. The legitimate approach is not manipulation but continuing to publish accurate primary information in a form that is easy to reference.
7. If you measure, control the conditions. Without recording the service, the model, whether web search was on, the language, the date and time, the history, the prompt, and the number of repetitions, you cannot isolate the cause of a change.

Next actions

Start by putting the questions a candidate is likely to ask about your company into several AI services. Two are especially high priority: "which companies in this industry are good places to work?" (does your company come up?) and "how does [your company] differ from its competitors?" (how you are positioned in a comparison). Record the measurement conditions from section 7.1 as you do it.

For most companies, a systematic and continuing capability for measuring how they appear on AI is likely still under construction. Noticing that students use AI is a starting point, not a destination.

For the full picture of what it means to grasp and manage perception on AI on a continuing basis, see What is AI Perception Management (AIPM)?.

Primary surveys and official documentation

1. Mynavi, "April 2026 Career Intentions Survey of University Students Graduating in 2027 (On AI Use Among Job Seekers)" (published May 26, 2026; fieldwork April 25–30, 2026; 1,258 valid responses)
   https://career-research.mynavi.jp/reserch/20260526_110798/
2. Mynavi, "2025 New Graduate Recruiting and Job Market Review (Class of 2026)" (published September 26, 2025; 30.6% of companies sensing student use of generative AI)
   https://career-research.mynavi.jp/reserch/20250926_101951/
3. Mynavi Career Research Lab, "Changes in Job Hunting Brought by Generative AI: From the Class of 2024 to the Class of 2025" (published August 9, 2024; the 12.3% figure as of the class of 2025)
   https://career-research.mynavi.jp/column/20240809_84095/
4. Google Search Central, "AI features and your website" (last updated December 10, 2025)
   https://developers.google.com/search/docs/appearance/ai-features
5. Consumer Affairs Agency (Japan), "Q&A on Stealth Marketing"
   https://www.caa.go.jp/policies/policy/representation/fair_labeling/faq/stealth_marketing/

Peer-reviewed papers

6. Tursunbayeva, A., Fernandez, V., Gallardo-Gallardo, E., & Moschera, L. (2025). Artificial intelligence and digital data in recruitment. Exploring business and engineering candidates' perceptions of organizational attractiveness. European Management Journal. DOI: 10.1016/j.emj.2025.03.002
7. Schilke, O., & Reimann, M. (2025). The transparency dilemma: How AI disclosure erodes trust. Organizational Behavior and Human Decision Processes, 188, 104405. DOI: 10.1016/j.obhdp.2025.104405

Surveys by interested parties (indicative values)

8. LANY Inc., "Survey on Silent Withdrawal Caused by AI in the Recruiting Process" (published June 4, 2026; fieldwork April 1–3, 2026; 111 valid responses; respondents aged 25 to 45 who had conducted a job search within the past year and researched company information with AI)
   https://www.lany.co.jp/lany-llmo-lab/job-candidate-ghosting
9. Nyle Inc., "Survey on the Use of Generative AI in Job Hunting" (published April 16, 2026; fieldwork March 25–30, 2026; 419 valid responses); source: research by Nyle's SEO Consultation Room
   https://www.seohacks.net/column/30459/
10. PerceptionX, "How Job Seekers Use AI to Research Employers" (May 2026; via Prolific; 306 respondents across seven countries; respondents were AI chatbot users who were also job seeking)
    https://www.perceptionx.ai/research/ai-candidate-usage
11. Al Ansari, K. "The invisible interview: How AI is reshaping employer brand," HR Executive (March 2026)
    https://hrexecutive.com/the-invisible-interview-how-ai-is-reshaping-employer-brand/

About this article

Every figure in this article was verified against primary sources as of August 2, 2026. This area changes quickly, so we recommend re-checking with the original publishers before citing.

AIO, GEO, LLMO, and AIPM are all practitioner terms rather than officially defined standards. Google's AI features are grounded in the same index and technical requirements as ordinary Search, and no AI-specific schema.org markup is required.