Pre-Edit Source Text With AI Before You Open Your CAT Tool

The Source Text Problem Nobody Charges For

Every experienced translator has had this day. The client sends you a 40-page source document. You open it. The first three paragraphs are fine. On page five, a bullet list suddenly uses inconsistent capitalization. On page eight, a term is spelled two different ways. On page twelve, a sentence is missing a verb because copy-paste went wrong in the client's review process. Page fifteen has a dangling reference to "the enclosed table" which is not attached. Page twenty has a section that looks like it was drafted by someone different, more formal, more jargon-heavy.

You are not paid to fix any of this. You are paid to translate it. But if you do not fix it, the translation inherits every problem the source had and the client blames you. So you burn 45 minutes cleaning the source up in your head or in comments before you can even start, on a 40-page project you are charging by the word.

The fix is obvious in theory and painful in practice: run a pre-edit pass before the source goes into your CAT tool. Most translators skip this because the pre-edit pass is just as tedious as the translation itself. An AI agent can take the tedious part off your plate and hand you a cleaner, ambiguity-flagged source file 10 minutes after you forward it.

What a Pre-Edit Pass Actually Involves

A proper pre-edit on a source document looks for the same handful of issues every time. Once you list them, it becomes clear why this is perfect for AI.

Inconsistent terminology. The same concept called "user" in one paragraph and "customer" in another.
Inconsistent capitalization and hyphenation. "eCommerce," "e-commerce," "Ecommerce" in the same document.
Spelling or grammar issues. Typos that survived the client's own QA.
Broken references. "See table 3" when table 3 is not attached, or numbering that skipped.
Ambiguous pronouns and subjects. "They will need to review it" where "they" and "it" are unclear.
Inconsistent voice. Sections written in different tones by different authors.
Layout fragments stuck in the text. "See attached file:" left over from a template.
Cultural and legal landmines. Dates in ambiguous formats, currency without codes, claims that do not translate cleanly into the target market.

A careful human can find all of these. A careful human takes an hour per 40-page document. An AI agent with a pre-edit prompt does it in under two minutes, flags each issue with a specific line reference, and asks you how you want to handle it.

How an AI Agent Runs a Pre-Edit Pass

The workflow fits in a single Telegram chat. Three steps.

1. Drop the source file in the chat

You forward the docx, the pdf, or the plain text to your Hermes agent. You can optionally add a short instruction: "This is going into French, marketing voice, 40 pages. Check consistency and flag anything unclear for translation."

2. Get the triage back

Within two minutes the agent returns a report like this:

Pre-edit triage, client file "brochure-v3.docx" (2,840 words)

Consistency:
- "user" (used 14 times) and "customer" (used 7 times) refer to the same concept. Decide on one and I will normalize.
- Capitalization of "Platform" is inconsistent, 8 uppercase, 3 lowercase. Your client's style guide says lowercase, I will normalize unless you say otherwise.

Ambiguities:
- Page 8, line 4: "they will review it" - unclear if 'they' is customers or the support team. Flag for query?
- Page 14, footnote: "See attached table" - no table detected. Check with client.

Spelling and grammar:
- Page 3, line 21: "acces" should be "access"
- Page 11, line 9: extra space before comma, cosmetic

Target market issues:
- Date format "03/04/2024" on page 6 is ambiguous, reads as March 4 in US and April 3 in most of Europe. Recommend writing out the month.
- Page 22 claim "FDA-approved" does not apply to EU market. Flag for client or localize to "CE-marked" if applicable.

Recommended next step:
Approve my suggested normalizations for consistency and spelling. Draft a short query list for the client on the 2 ambiguities and 1 market issue. Then we can open the CAT tool with a clean source.

3. Approve and go

You read the report, approve the normalizations, copy the query list into an email to your client for quick clarification, and open Trados or memoQ with a source that no longer has the 11 issues listed. By the time the client replies to your query email, you are 30 percent through the translation.

What This Changes for Your Working Day

Three concrete outcomes.

Fewer rework cycles

Most translation rework happens because an ambiguity in the source got resolved the wrong way. A pre-edit that surfaces the ambiguities early means you asked the client before translating, not after delivering. This alone cuts rework by a large margin over a year.

Faster first pass

A clean source is easier to translate. Segments are clearer, terminology is already normalized, typos are gone. Your translation speed in words per hour goes up without you working harder.

Better client perception

When you email the client with five specific, numbered queries about their source before you start translating, you look like a serious partner rather than a vendor. The queries are easy to answer because they are precise. Over time, clients start sending you better source documents because they know you will notice.

Tuning the Pre-Edit to Your Style

Every translator wants slightly different things flagged. The agent learns.

Tell it what to ignore. "Do not flag number formatting inconsistencies, I will handle those in the CAT tool with a find-replace." The agent stops flagging them.

Tell it what to flag more aggressively. "Any sentence with more than 30 words, flag for potential restructuring." It adds that.

Feed it your style guide. "Apply this style guide as the baseline for all consistency checks." It becomes the default lens.

Save different pre-edit profiles per client. Client A wants heavy flagging. Client B just wants typos. You say "run Client A pre-edit" and the right profile runs.

What the Pre-Edit Is Not

Three things it is not trying to be.

Not a translation. The agent is not translating the source. It is preparing it. You still open the CAT tool and do the real work.

Not a substitute for human judgment. If the agent flags something as ambiguous and you know from context that it is fine, ignore it. Flag means "worth a look," not "must be changed."

Not a replacement for clarifying with the client. The pre-edit generates a query list. The query list still needs to go to the client. Machines do not resolve ambiguity, people do.

The Case For Charging For It

A practical business note. If you regularly pre-edit for clients, you are doing work that has economic value to them. Many agencies have a separate rate for pre-edit services at about 50 to 70 percent of the translation rate per word. With an AI agent running the pass for you, the margin on this becomes substantial.

You have two options:

Include a free "source review" pass as part of your standard service, raising your per-word rate slightly to cover the time. Most clients prefer this because it is one invoice and one deliverable.
Offer pre-editing as a line item. Charge 40 to 60 percent of your translation rate. Deliver the cleaned source as a separate deliverable alongside the translation.

Both work. Option 1 is typical for direct freelance clients. Option 2 is typical for agencies that are used to line-item pricing.

Getting Set Up in One Evening

Spin up a Hermes agent on Hermify. Sixty seconds.
Load your style guides and a couple of example pre-edit reports you would write if you had infinite time. The agent learns the voice and depth you want.
Run the pre-edit on your last finished project as a test. Compare what the agent flags to what actually caused rework. Tune from there.
Add the pre-edit step to your project onboarding workflow. Every new file goes through the agent before it goes into your CAT tool.
Track the time saved for one month. Use the numbers in your next rate negotiation.

The Larger Point

Translators compete on speed, quality, and reliability. A pre-edit pass is the single highest-leverage change most freelancers can make to all three at once. The only reason more translators do not do it is that the manual version is as tedious as the translation itself. An AI agent removes the tedium and keeps the upside.

You already know what a clean source does to your work. Start every project with one.