Resume Formatting AI Tools for Recruiters: Check Nothing Was Altered

Recruiter-side formatters such as Candidately, FormaCV and CVFormatter reflow a CV into your agency template for roughly $0.60 to $1 a CV, while Cluit rewrites content into a pack from $29 a month. To check nothing was altered, run a word-level comparison against the original, then verify every date, title and employer.

Search results for this topic mix two kinds of product: recruiter tools that put a CV into a template, and job-seeker builders such as Rezi and Kickresume that exist to improve a candidate's wording. The first kind can leave content untouched. The second kind is built to change it. Even among recruiter tools, several offer optional rewriting, summarising or tailoring to the job, so whether a CV comes back unchanged depends on settings you have to check on every run.

Follow me on Instagram@sagnikteaches

Reflow, rewrite and job-seeker tools are different products

Before comparing prices, sort each tool by what it does to the candidate's words. The vendor pages I read on 1 November 2026 say this:

Connect on LinkedInSagnik Bhattacharya

Reading the "what it does to the words" column is the whole decision. A job-seeker tool in your agency workflow would change the candidate's CV on purpose, and an agency that sends that version to a client is vouching for words the candidate never wrote. Cluit's own pages say it rewrites the resume and its bullets, so it is a rewriting product with a formatting side. CVFormatter makes the cleanest promise, that content stays as written unless you pick an AI action, but a promise on a pricing page is not a test result. FormaCV and Candidately both list wording changes as features, and with those the safest assumption is that the default may change words until you have shown otherwise.

Subscribe on YouTube@sagnikteaches

Several of these tools also summarise. That is a separate job with its own risks, and the tutorial on writing candidate summaries clients read covers it. Keep the two apart: a formatter should not summarise, and a summary should not be hidden inside a formatted CV unless the candidate and the client both know it was written by software.

What each tool costs per CV, and why checking costs more

List prices turn into a cost per CV with one division. These figures assume you use every credit you pay for, which is generous, because CVFormatter's monthly credits do not roll over.

Now take a six-consultant agency that formats 300 CVs a month. CVFormatter Premium plus one $49 pay-as-you-go pack of 50 covers 300 for $228. FormaCV comes to $297 and Candidately to $300. Cluit's Agency plan is $79, but it is the one that rewrites. So the tool bill ranges from $228 to $300 for the reflow tools, a spread of under $75 a month.

Compare that with the check. At three minutes a CV, 300 CVs is 15 hours a month, which is $450 if consultant time is worth $30 an hour. The check costs more than the most expensive tool, so it is the number to manage. Every alteration you find adds minutes, so the cheaper tool that changes words more often is not the cheaper tool. Choose on how rarely a tool alters content in your own tests, which the canary CV later in this tutorial measures, and use price to break ties.

The unaltered-content check, in six steps and about three minutes

1. Keep the original next to the output. Never overwrite it. Name the files with a reference number rather than the candidate's surname, for example REF1043-original and REF1043-blind, so the blind file's name does not give the candidate away.
2. Compare the words. In Word, open the Review tab, find the Compare group, choose Compare, then "Compare two versions of a document (legal blackline)". Put the original under "Original document" and the formatted CV under "Revised document", set "Show changes" to word level, and leave the result going to a new document, which is the default, so neither file is touched. If the original is a PDF, copy its text into a blank Word document first. Expect noise from line breaks and bullet characters; read only the inserted and deleted words.
3. Sort the changes into three piles. Expected: reordered headings, bullet characters, and contact details removed on a blind CV. Explainable: a date format tidied, with nothing lost. Not acceptable: everything else.
4. Check the facts one by one. For each role: employer, title, employment type (permanent, contract or part-time), start and end. For education: qualification, institution, classification and dates. For numbers: every figure with a unit. The spreadsheet routine below makes this quick.
5. Count things. The number of roles, education entries and bullets should match. The first and last roles must both be present. A missing final role is the classic sign of a tool that ran into a length limit. Total words should sit within about 5 percent of the original unless you removed something on purpose.
6. Log it. Record the reference, date, who checked, the tool and whether its AI options were on, and the number of alterations found. If a client later questions a date, the log is your evidence that someone looked.

For step 4, type the facts from the original into column B and the same facts from the formatted CV into column C, then put this in column D:

=IF(EXACT(TRIM(B2),TRIM(C2)),"OK","CHECK")

EXACT is case sensitive, so a quiet change from "Assistant" to "assistant" or "Ltd" to "ltd" also shows up. If you would rather not retype, paste both versions' lines of dates and titles side by side and let the formula do the same job.

Here is what a filled-in check looks like for one invented CV, again with invented details. Word's comparison showed nine changes; the fact check classified them like this:

Six of the eight lines are alterations. Three would be noticed by a client reading the CV against an interview, and the invented "Present" could cost you the client. The fix is mechanical: restore the original wording, rerun with the tool's AI options switched off, repeat the comparison, and log what you found. If the same tool repeats the pattern on the next two CVs, the tool is not fit for clean reflow, whatever its page promises.

Where AI tools quietly alter CVs, and how each shows up

Alterations are rarely dramatic. They are small tidy-ups that read well and are wrong. These are the families to look for, with the quickest way to catch each:

One more risk belongs on the list even though it is not an alteration: a tool that adds a summary or a "red flags" analysis produces content the candidate never saw. Cluit's pack includes a red flags analysis, Candidately lists AI candidate summaries, and FormaCV's tailoring can bold some details and remove others it judges irrelevant. Decide whether you want that content in anything that leaves your agency, and whether the candidate has agreed. The tutorial on candidate records, duplicates and follow-up in a recruitment CRM covers where such output should be stored.

A canary CV to test any tool before you buy

A trial on a real candidate puts a real person's record at risk. Instead, build a fake CV with invented names and plant traps in it. Twelve are enough:

1. A role that starts "03/04/2021" and ends in a named month, to test date handling.
2. A contract role and a part-time role.
3. A three-month role sandwiched between two long ones.
4. A four-month gap with no role at all.
5. Two overlapping roles.
6. An employer with a legal suffix and a "(formerly ...)" name.
7. A master's degree "with distinction".
8. A qualification marked "2018 to 2020, not completed".
9. A number with a unit, such as "managed 7 staff and a $1.2m budget".
10. A name with diacritics and a double-barrelled surname.
11. A skills list with an unusual tool and an acronym in capitals.
12. Three pages, with the last role on page three, to test length.

Run the canary through each tool's free plan or trial twice: once with every AI option off, and once with them on. Compare each output with the original using the six-step check and count the traps that changed. A made-up scoreboard shows how to read the result:

Tool A is the only one that meets the bar, and only with AI off, so the buying decision is to use it as a reflow tool and keep its AI options disabled. Tool B fails at default settings and Tool C looks fine until someone ticks an option, which is exactly the kind of difference a pricing page never shows. Keep the canary file. Run it again after any notice that the vendor has updated its models or templates, because the output you tested may not be the output you now get.

Blind CVs: what to strip and how to prove it is gone

A blind CV removes what could identify a candidate before a client sees it. The usual list is the name, email address, phone number, home address, photo and profile links (LinkedIn, portfolio, social accounts). Many agencies also remove dates of birth, marital status and nationality where a CV includes them. Employer names are a judgement call. Some clients cannot assess experience without them; others accept a descriptor such as "a 200-person logistics firm". CVFormatter offers to hide employer names or replace them with an industry description, and Candidately hides previous employer names with one click, so decide per client rather than per tool.

Removing is half the job. The other half is proving nothing is left, because identifying details hide in odd places. Run this leak check on every blind CV, and run it on the canary first:

• Search the output for the surname, the first name, the local part of the email address, the last four digits of the phone number, the LinkedIn handle and every employer name you meant to hide.
• Search for two distinctive project or product names from the original, because a project name can identify an employer as surely as the employer's name does.
• Look in headers, footers, text boxes and hyperlinks. A link whose visible text says "Profile" may point to a page with the candidate's name in the address.
• Check document properties. In Word, open File, Info, Check for Issues, Inspect Document and remove personal information, running it on a copy because removed data often cannot be brought back with Undo. For PDFs, open the file's properties and check the author and title fields.
• Check the file name and any share link. CVFormatter says its share links expire after 30 days, but a link to the original is still a link to the original.

Consent comes before any of this. Tell candidates, at registration and again before a CV goes to a client, that software will reformat their CV and strip identifying details. Something like this, adapted to your own terms, works as a registration line:

We use software, which may include AI tools, to put your CV into our template and, when a client asks for it, to remove your name, contact details and employer names before we share it. The software is not meant to change what your CV says, and we check the result against your original. [Tool name] keeps your CV for up to [period]. Tell us if you would rather we did this by hand.

Then check what each vendor says about the data. CVFormatter's page says uploaded CVs are deleted automatically after 30 days, that your team can delete any CV sooner, that candidate data is not used to train AI models, and that data is encrypted in transit and at rest, and it mentions a data processing agreement. FormaCV says customer CVs are never used to train AI models and lists an audit log of every access, edit and download, with data-residency regions and an on-premises option for enterprise customers. Candidately lists GDPR compliance, encryption at rest and a SOC 2 Type II audit it describes as in progress. The Cluit pages I read say nothing about retention or training. These are vendor statements, so ask each for its data processing agreement, its list of sub-processors, its retention period and its position on training before you upload a real CV. If you have candidates in the EU, data-protection rules expect you to tell them how their CV is processed and to have a lawful basis for it, so have a qualified adviser look at your registration wording.

The EU AI Act adds a second question. Recruitment is on its high-risk list in Annex III, with obligations for stand-alone systems deferred to 2 December 2027. A tool that only reflows text is not obviously in that category, but one that summarises, highlights or ranks candidates moves closer to it. The tutorial on what high-risk means for a recruitment agency explains where the line falls, and it is a reason to keep a formatter doing only formatting.

Keeping a general AI assistant to reflowing

You do not need a dedicated tool to reflow a CV. A general assistant can do it if the prompt forbids editing and you verify the output. The plan matters: use a business plan, because candidate CVs are personal data, and not a personal free account. Claude Team's pricing page states no model training on your content by default; other vendors' business plans make similar statements, so check the one you hold. Here is a prompt to start from:

You are a CV formatter, not an editor. Reformat the CV below into this
structure: Name line, Profile (only if the original has one), Experience
(reverse chronological), Education, Skills, Other.

Rules:
1. Do not add, remove, reword, shorten, merge or "improve" any text.
2. Keep every date, job title, employer name, qualification, number and
   unit exactly as written, including brackets such as "(contract)".
3. Never add "Present" or any date that is not in the original.
4. If something does not fit the structure, put it under Other, unchanged.
5. If text is unclear or cut off, write [CHECK: describe it] and leave it.
6. After the CV, list every place where your output differs from the
   input, or write "No differences".

CV:
[paste the candidate's CV text]

The last rule invites a false sense of safety. An assistant that reports "No differences" can still have smoothed a date, because the report is generated by the same process that made the change. Treat it as a useful prompt for your own check, never as the check. In one illustrative run the assistant replied "No differences" and Word's comparison then showed three: "contract" removed from one title, "Aug 2021" shortened to "2021" and a trailing "(formerly ...)" dropped. That is why the six steps above run on every output, however it was produced.

How often to check once a tool earns your trust

Checking every CV for ever is not realistic, and a sampling plan is better than abandoning the check. A workable pattern is to check every CV for the first 50, then one in five plus any CV with an unusual layout, such as a two-column design, a career break or a long list of short contracts. Return to full checks for the next 20 CVs after any vendor update, any new template, and any alteration you find.

The saving is real. At 300 CVs a month, checking all of them at three minutes each takes 15 hours. Checking the first 50 in full and then a fifth of the rest takes 100 CVs, which is five hours, so the sampling plan saves 10 hours, about $300 at $30 an hour. Spend a little of that on a containment rule: when a client or candidate reports an altered fact, pull every CV formatted in the last 30 days with the same template and settings, run the comparison on each, and correct and resend any that changed. It will take an afternoon once, and it protects the relationship far better than a reply that says the tool must have made a mistake.

Keep the log where the team can find it, in the applicant tracking system if it supports notes or version history (Candidately lists profile version history as a feature) or in a shared sheet if not. A client rarely asks for proof that a CV was checked. When one does, a log with a date, a name and a count is the difference between an answer and an apology. The same consent thinking applies to meeting recordings, covered in the tutorial on candidate consent and tool choice for AI note takers.

Questions recruiters ask about AI CV formatting

Do I need the candidate's consent to anonymise a CV?

Tell candidates, at registration or before you send, that their CV will be reformatted by software and have identifying details removed for the client, and keep a record of their agreement. The legal basis depends on where you and the candidate are, so ask a qualified adviser to confirm your wording.

Can I use ChatGPT or Claude to format CVs?

Yes, on a business plan, with a strict reflow-only prompt and a comparison afterwards. Claude Team's pricing page says it does not train on your content by default; confirm the same for any other plan you use. A general assistant tends to smooth wording unprompted, so never skip the word-level check.

How often should I check formatted CVs?

Check every CV for the first 50, then a sample of one in five plus any CV with an unusual layout. Return to full checks for 20 CVs after any vendor update, any new template or any alteration you find. The check takes about three minutes once you have the routine.

Which tool is cheapest per CV?

On list prices, CVFormatter's largest plan works out near $0.60 a CV if you use all 500 credits, and FormaCV and Candidately sit near $1. Cluit is cheaper per pack but rewrites content. Your checking time, about three minutes a CV, usually costs more than the tool.

What if the comparison shows dozens of changes?

Sort them first. Layout noise, reordered headings and removed contact details in a blind CV are expected. Any change to a date, title, employer, qualification or number is not. If a tool alters facts on a clean reflow, switch its AI options off, retest with a canary CV, and move vendors if it persists.

Further reads

• AI Staffing Best Practices: A Checklist for Placements and Candidate Data — A checklist for placements and candidate data
• Is AI CV Screening Fair? Bias Checks Every Recruiter Should Run — Bias checks if your tool also ranks candidates
• How to Evaluate the AI Features in Your Recruitment Software — How to judge AI features before you buy
• ChatGPT Prompts for Recruiters: Intake, Outreach and Shortlists — Prompts for intake, outreach and shortlists
• AI CV Screening for Recruitment Agencies: Setup and Safeguards — Safeguards if you also screen CVs with AI
• AI Sourcing Tools for Recruiters: Seat Costs and Consent Rules Compared — Five sourcing tools priced for three recruiters, with data sources, consent handling and a one-live-search trial you can run in a week.
• AI Prompts for Career Coaching: A CV and LinkedIn Review Workflow — A career coach's review workflow: consent, intake form, six prompts with a before-and-after example, and the judgements that stay with the coach.
• AI Tools and AI Development: The Complete 2026 Guide — the AI hub, including every tutorial in the AI-for-business series.

Sources: vendor pages for Candidately, FormaCV, CVFormatter, Cluit, Rezi and Kickresume, and Microsoft's help pages for Word's Compare and Document Inspector, read 1 November 2026; Claude Team pricing page; EU AI Act Annex III. The CV examples are invented.