Research programme of Elite Careers StrategyExplore evidence · compare
Compare AI rules by selection stage
Of the 8 employers in our first AI-rules study, 6 prohibit candidate AI in at least one online test or interview, and 3 state no interview rule in the reviewed official sources (observed 23 September 2026).
Population: the first eight-employer AI-rules study, Eight institutions in the current research corpus for which an official current source explicitly addresses candidate AI use or AI/human decision controls: McKinsey & Company, UBS, PwC UK, Clifford Chance UK, Deloitte UK, HSBC, BCG and A&O Shearman UK. Each record is read from the AI Rules Tracker v2.0. See also finding WGI-FIND-001 and the method and limits.
Stage by stage
| Stage | McKinsey & CompanyGlobal | UBSGlobal | PwC UKUnited Kingdom | Clifford ChanceLondon | Deloitte UKUnited Kingdom | HSBCGlobal | BCGGlobal | A&O ShearmanLondon (guidance also listed for Singapore, Hong Kong) |
|---|---|---|---|---|---|---|---|---|
| Application materials | EncouragedWhen used responsibly, we see AI as a valuable tool that supports preparation. Misuse, however, compromises fairness.Official source | Permitted with limitsIt's ok to use AI tools to help prepare your application documents (like your resume or cover letter), but the content must genuinely reflect your own skills, experience, and understanding.Official source | Not statedNo statement located in the reviewed official sources. This is not permission. | Permitted with limitsWhile AI tools can support your preparation, overreliance may weaken your application.Official source | Permitted with limitsYou may use AI feedback tools to practice for interviews, although your responses must be based on your own experiencesOfficial source | Permitted with limitsHelp structure your CV or application, provided the content accurately reflects your own experience and skills.Official source | Permitted with limitsAI can support you in structuring your responses, identifying key themes, and improving clarity.Official source | Permitted with limitsAI can help with the formatting and design of your CV and help tailor it to the role you're applying for.Official source |
| Online tests | Prohibited unless authorisedUsing AI during assessments (unless specifically permitted)Official source | ProhibitedYou must complete all assessment tasks independently - AI assistance is not allowed.Official source | Prohibited unless authorisedThe use of AI tools, AI-enabled services, or any third-party assistance to generate, edit, rewrite, summarise, coach or otherwise complete assessment responses is not permitted unless explicitly authorised by PwC.Official source | ProhibitedPlease note that the use of AI is not permitted.Official source | ProhibitedCompleting online assessments, including the Immersive Online Assessment, Onestop and Job Simulation: these assessments must be completed independently and without AI assistance.Official source | General authenticity onlySubmit work that you cannot personally explain, validate, or replicate if asked during the recruitment process.Official source | Not statedNo statement located in the reviewed official sources. This is not permission. | General authenticity onlyWe simply ask that your application and assessments reflect your own thinking, experience and abilities.Official source |
| Interviews | ProhibitedGenerating real-time answers during an interviewOfficial source | ProhibitedDuring recorded or virtual video interviews, your answers must be your own. Using AI to generate responses is strictly prohibited.Official source | Prohibited unless authorisedThe use of AI tools, AI-enabled services, or any third-party assistance to generate, edit, rewrite, summarise, coach or otherwise complete assessment responses is not permitted unless explicitly authorised by PwC.Official source | Not statedNo statement located in the reviewed official sources. This is not permission. | ProhibitedInterviews must be completed without AI assistance.Official source | ProhibitedGenerate responses during live or recorded interviews.Official source | Not statedNo statement located in the reviewed official sources. This is not permission. | Not statedNo statement located in the reviewed official sources. This is not permission. |
| Final-stage tasks | Not statedNo statement located in the reviewed official sources. This is not permission. | Not statedNo statement located in the reviewed official sources. This is not permission. | Not statedNo statement located in the reviewed official sources. This is not permission. | Not statedNo statement located in the reviewed official sources. This is not permission. | Sanctioneduse GenAI tools, such as ChatGPT, Gemini or Microsoft Copilot, alongside other research methodsOfficial source | Not statedNo statement located in the reviewed official sources. This is not permission. | Not statedNo statement located in the reviewed official sources. This is not permission. | Not statedNo statement located in the reviewed official sources. This is not permission. |
| Employer use of AI | Discloses AI useThe interview you are about to undertake is conducted entirely by an AI system, specifically, an AI-powered chatbot.Official source | Not statedNo statement located in the reviewed official sources. This is not permission. | States AI not usedWhile PwC uses AI to support aspects of its business, we do not use AI to assess candidate performance within the Early Careers selection process.Official source | States AI not usedWe do not use AI to screen applications or make hiring decisisons [sic]Official source | Not statedNo statement located in the reviewed official sources. This is not permission. | Discloses AI useNo, the AI-based platform does not make any automated decisions in the hiring process.Official source | Discloses AI useAI-assisted screening tools can highlight experience and potential but never make decisionsOfficial source | States AI not usedWe do not use AI at any point during our recruitment process; all assessments and decisions are made by our team.Official source |
| Hiring decision | States humans decide onlyNo candidate is rejected solely on the basis of the chatbot's output. All AI-generated assessments are reviewed by a human recruiter before any decision is made regarding your candidacy.Official source | States humans decide onlyWorried employers are using AI to review submissions? Rest assured; our people are the ones making decisions around here.Official source | States humans decide onlyInstead of having an interviewer, your answers will be recorded and reviewed afterwards by one of our assessors.Official source | States humans decide onlyyour application will be reviewed by a member of the team once you have passed the online assessments.Official source | Not statedNo statement located in the reviewed official sources. This is not permission. | States humans decide onlyAll decisions will be made by our hiring team.Official source | States humans decide onlyAI is never used to make hiring decisions at BCG. All decisions are made by experienced BCG recruitersOfficial source | States humans decide onlyWe do not use AI at any point during our recruitment process; all assessments and decisions are made by our team.Official source |
Relation to WGI-FIND-001 The finding counts seven protected-stage policies and states that they are not seven equivalent bans. This table uses the stage codes from the 23 September re-check, where A&O Shearman's assessment rule is coded as a general authenticity requirement rather than a prohibition.
Reading this table "Not stated" means no statement was located in the reviewed official sources. It is never a permission. A prohibition applies to the stage and programme its source names, not to every route at that employer.
Deadlines by institution
Not yet published. Deadline signals do not yet carry a structured date field, so a timeline would have to read dates out of free text. That is withheld until the dates are recorded as data.
What this does not show
- N=8 is a defined evidence subset, not a prevalence estimate for elite employers.
- Policies are institution-, market-, programme- and stage-specific and may change.
- Absence of a coded permission or prohibition is not evidence that no such policy exists.
- A general authenticity requirement is not automatically coded as a prohibition on AI during every assessment stage.
- The seven protected-stage policies are not seven equivalent AI bans. A&O Shearman specifies candidate-owned thinking in assessments; Clifford Chance explicitly prohibits AI in Watson Glaser. Policy publication does not establish enforcement or hiring effects.
- Codes come from the AI Rules Tracker v2.0 re-check; each cell is bounded to the market and programme scope of its source.