AI in Consulting Recruiting: Where the Firms Disagree

AI in consulting recruiting has no industry rule

Reading one firm's page teaches you almost nothing about the next one. The cleanest proof is an object nobody argues about.

McKinsey's integrity page bans "using any applications or websites, generative AI, a calculator, or prepared notes" in any assessment or interview. Bain's digital assessment page allows one: "A pen, paper, and calculator are allowed." Same year, same kind of online screen, opposite rules on the same device. Bain draws its line around AI. McKinsey draws it around every external aid, arithmetic included. Practicing on a calculator because one firm allows it will cost you in the other firm's interview, where the working is the graded part.

The spread is also lopsided. Twelve of these firms publish something about what you may do with AI. Four publish something about what their own software does to your file, and two of those four, OC&C and L.E.K., publish only that half. The screening disclosures rarely sit on a careers page at all. OC&C puts its in a job advert. L.E.K. and Roland Berger put theirs in a recruitment privacy notice, which almost nobody opens.

Dates matter here more than on most topics. Only one of these documents carried a visible date when we read it in October, McKinsey's Solve FAQ, marked August 2025. The rest carry none, so we record the day we read each one. Advice written before this year describes a different set of rules.

Every firm's published rule on your AI use

One row per firm, quoted from the page linked in the last column. A blank in the permitted column means the firm names no permitted use, which is silence rather than a ban.

Three readings fall out of that table. PwC's ban is the widest in the set, because it reaches the verbs edit, rewrite and coach, so a grammar helper is out for assessment answers. Deloitte publishes two different expectations in two markets, and the UK one is the only instruction anywhere to go and use a named chatbot. The last two rows name an ability the firm wants while granting no permission inside an exercise, and an absent prohibition will not work as a defense. Our BCG guide, Deloitte guide and Simon-Kucher guide cover the rounds these rules sit inside.

The three AI rules that hold at every firm

Every firm that addresses the question answers these three the same way. Treat them as the floor, and treat the rest of the table as local.

Never generate an answer in real time during an interview. All five firms with a published position say it. IBM publishes the scenario in plain words: "Having ChatGPT open during a Zoom interview to suggest answers in real time."

Never run AI during an assessment. Bain's wording is the broadest of the set, because it covers "any software that transcribes or provides feedback during the assessment." A meeting assistant is caught by that sentence as surely as a chatbot. Deloitte UK bans AI across its Immersive Online Assessment, Onestop and Job Simulation by name, which is the only place a firm lists the assessments it means.

Disable installed AI software before any video call. This one catches honest people. A meeting assistant set up for your own work can join a call uninvited. McKinsey states it directly: "Disable previously installed AI software (such as note-taking tools or virtual assistants), as these may unintentionally record or assist with the interview."

Two smaller rules travel with these. Do not record or screenshot an assessment, which McKinsey and Bain both write out. And do not describe the content afterwards, which McKinsey states and PwC checks for by looking for "identical transcripts."

What happens if a firm decides you used AI

The consequences are published far more often than the detection methods, and they reach well past a rejection.

Accenture states that it looks: "We use proctoring tools and AI-detection methods to ensure a level playing field." Deloitte UK says "we have measures in place to detect AI-generated or AI-assisted responses." Neither firm names a tool, a vendor or a method, and no other firm here says it detects anything at all.

Two of those reach outside the hiring process entirely. Deloitte's survives your start date, and PwC's reaches your university. They are the strongest statements in the set, and they are the reason the grey middle deserves attention rather than a shrug.

KPMG's is the mildest and the most usable, because re-testing is a second sitting rather than a rejection. Two firms also make you say the words first. KPMG asks you to confirm the rules before each assessment, and Deloitte UK asks for "a short integrity statement confirming your understanding and agreement to these guidelines on camera." No firm publishes an appeal route for a candidate flagged wrongly, and that absence is the largest hole in the published record. Our PwC guide and KPMG guide cover the stages around these checks.

AI in consulting recruiting also reads your CV first

Now turn the page over. Accenture publishes a hiring process AI FAQ explaining what its screening tool does to your application, and it is the most specific document of its kind in this set.

The mechanism is plain: "This solution creates a grade by comparing the information a candidate provides during the application process against the required and desired qualifications of the open role. A grade is created for each candidate (A, B, C, or D)."

The FAQ then answers the question a candidate has. You are not ranked against other people: "The grade represents how well the candidate information matches with that of the requisition." It is recalculated each time you apply to a new role, and the firm writes that it "is not used to determine who to hire."

Four rights come with it, all published:

The vendor is named as well: "Hiredscore, a part of the Workday organization, works on Accenture's behalf." The candidate page repeats the choice in one line: "You can choose whether you consent to AI being used as part of your evaluation."

One practical move follows from how the grade is built, since it compares your application against the posting's own words. Where a posting names a skill you genuinely have, write it the way the posting writes it. The Accenture Strategy guide covers the case format after this step.

Where a decision is made before a person looks

Two firms disclose that something happens to your application with no human in front of it, and they handle it very differently.

OC&C puts its disclosure in the job advert. The London Associate Consultant vacancy says: "Some CV screening decisions may be made using the tool without prior individual human review, subject to appropriate safeguards and human oversight, however, final recruitment decisions are made by people." The tool is described as "an internally developed AI tool to support our recruitment team in reviewing applications."

Sit with the middle clause. Some screening decisions happen before any person sees your file. The firm names safeguards without describing them, and offers no consent step and no opt out. The same advert says "Please remove name & contact details," so what carries you is the content of each line. Our OC&C guide sets out the six steps this sits above.

L.E.K. goes further and is more generous about it. Its recruitment privacy notice says the firm "may use automated decision making." For consulting roles in its London offices the trigger is academic: a 2:1 undergraduate result, and AAB at A level with an A in mathematics where taken. Below that, "you will automatically receive an email from us letting you know that your application has not been successful." Then the sentence nobody else here publishes: "You have a right to obtain human intervention and express your point of view and contest the decision." An address is given for doing it.

Three firms publish commitments pointing the other way. BCG is the most categorical: "AI is never used to make hiring decisions at BCG." Its tools "can highlight experience and potential but never make decisions or eliminate candidates." PwC UK says it does "not use AI to assess candidate performance" in Early Careers. Roland Berger's advert promises that "Every application is reviewed by a member of our (human) recruitment team." Its own privacy notice then describes natural language processing that reviews the job description, summarizes your application, and calibrates criteria against it, while stating that "there is no automated hiring decision." Both texts are official and neither is false. Software reads and summarizes, a person decides. A careers page and a privacy notice can describe one process in tones that do not match, and the privacy notice is usually the more literal of the two. Our L.E.K. guide and Roland Berger guide cover the rounds behind these filters.

When the AI is the one asking the questions

A third shape exists that neither the bans nor the screening rules cover. The AI sits on the other side of the table, and three firms here publish that it does.

KPMG's UK application process page is unusually direct: "You will be presented with 6 pre-recorded questions asked by our AI avatar Leah. You will then be given a set time to prepare and then record your responses. Although your interviewer is AI generated, we'd like to reassure you that your recorded answers will be reviewed and scored by a member of our Student Recruitment team."

The reassurance is the load-bearing half. An avatar presents, and people mark. If that split holds, nothing about your delivery should change because the face is synthetic.

McKinsey describes its own version on the assessment side. Its Solve FAQ says the assessment uses "advanced AI methods to ensure that there is ample variation in each scenario, and it is designed so that no two individuals have the same parameters and combinations of data." That has a direct consequence for how you prepare. Comparing notes with someone who sat Solve last week tells you nothing reliable about what you will see. Our McKinsey guide covers the rest of that process.

Accenture states the same category in one line: "Some roles may involve assessments that use AI to help us better understand your skills and potential." It does not say which roles or which assessments. What all three share is that your conduct rules do not loosen because the counterpart is software. Every ban above applies to a recorded interview exactly as it applies to a person on a call.

A checklist for the day before, in order

None of this is clever. It is the set of actions that satisfies the strictest published rules, which means it satisfies all of them. Steps two and three come from McKinsey's assessment integrity expectations, which reads like a setup checklist rather than a policy.

1. Read the conduct paragraph in your invitation email. It overrides every general page, including this one, and it is the only document written for your specific step.
2. List what is installed on the machine you will use. Note takers, meeting assistants, browser copilots, coding copilots, grammar helpers. Quit them and remove their calendar access, because some join calls on their own.
3. Disconnect the second monitor. McKinsey asks candidates to "only use one screen without overlays." No firm asks for the opposite.
4. Turn off virtual backgrounds and blurring. McKinsey asks you to "be on camera and disable any virtual backgrounds or blurring features." A plain wall is fine.
5. Reposition the camera to "show your full upper body and hands as you take notes and perform calculations." Check your own video feed before the call rather than during it.
6. Put pen and paper within reach, and put the calculator away unless the firm allows it. Bain allows one on its digital assessment and bans it in later interviews. McKinsey bans it throughout.
7. Sit somewhere nobody will speak. McKinsey asks you to "participate alone, without assistance from others including impersonation," and every firm that addresses the question agrees.
8. If you used AI heavily on something you submitted, say so in a line. IBM asks for exactly that, and no firm penalizes the honest version.
9. Afterwards, keep the content to yourself. McKinsey asks you to "discard any notes taken during the assessments/interviews" and to share nothing about the questions.

Do steps two and three the evening before, because an uninstall takes longer than it looks. One more detail from that page is worth reading twice: McKinsey "may ask to review your notes during the interview," so your working is visible while the case runs. Practicing on a keyboard trains a habit you will not be allowed to use.

What a structure is scored on when no tool can help

Once the tools are off, the interview measures the thing these rules exist to protect. The hundred points below are ours, written by us. None of the thirteen firms on this page publishes a rubric, a dimension or a weight, so nothing here is borrowed from one.

What none of these firms publish about AI

How detection works. Accenture and Deloitte both say they look. Neither names a tool or a method. The rest say nothing at all.

What happens if the detection is wrong. No firm publishes an appeal route, a review step, or a contact for a candidate who believes a flag was mistaken. L.E.K.'s right to contest is the closest thing here, and it covers an academic threshold rather than a judgment about AI.

How often anyone is disqualified. The consequences are published. The numbers are not, by anyone.

What "specifically permitted" covers at McKinsey. The exception exists in the table, and no published example of it does. Treat it as inactive unless your invitation email activates it.

Where the McKinsey attestation happens. The page refers to "attesting to these expectations" without saying at what point, or what you sign.

What OC&C's safeguards are. The word is published. The content is not.

Whether any of this survives a border. Deloitte already publishes two different rules for the UK and the US, and PwC's promise covers UK Early Careers only. Assume the rule is local until the firm says otherwise.

What the quiet firms do. We found nothing on either half from Oliver Wyman or Arthur D. Little, and Kearney serves its careers pages in a way we could not read at all. Silence is neither a ban nor a license. Where a firm publishes nothing, the authority is the email that invited you, and the recruiter who sent it can answer a direct question about tools without it counting against you.

What changed on this AI recruiting page, and when

7 October 2026. Every one of the twenty sources behind this page was opened again, and two findings changed what it says. Bain's digital assessment page now states that a calculator is allowed on the assessment and banned in the interviews that follow. Our 29 September reading recorded only the first half of that, so the calculator comparison with McKinsey now carries the restriction. And two dates we had printed in the body, 19 May 2026 on McKinsey's integrity page and 18 March 2025 on Accenture's screening FAQ, were not visible in the renderings we could read in October. Neither page is served to us directly, so we cannot tell whether the date moved or whether the proxy dropped it. The page therefore no longer asserts either date, and the registry entries say why. Everything else held word for word, including every quotation.

The same day the page was shortened by close to half. Eight sections that each handled one firm became the single table of twelve rows above. No published rule was dropped in that move, and every quotation kept its source.