How do you write a CV?

Most of the time spent writing a CV goes into deciding rather than writing: should I include that internship, how should I give the dates, how many tools should go in the list, how do I describe that job. Every one of those decisions has a consequence, and the consequence does not live in personal taste, it lives in what the side reading your CV can verify.

In GoTeam CV analysis the same fixed list of 49 items is applied to every CV. The items do not ask is it good, they ask is this written in this CV; 45 of them are computed directly in code. A good CV is open to argument, did you write the date as month and year is not.

The five steps below are built on that distinction. Each step makes one writing decision, says which item that decision satisfies and with what weight, and writes down what is lost when the decision goes the other way. No step says your score goes up by this much: score gains were not measured, the only thing measured is which item looks at what.

• The weights are within a dimension. A weight of 12 does not mean 12 out of the overall score, it means 12 units out of that dimension budget of 100 units.
• An item with no evidence does not count as half a point, it leaves the denominator. The absence of evidence is never counted against the candidate.
• The rubric is versioned: if the text or the weight of an item changes, the version goes up and an old score is not compared with a new one. The numbers here are the numbers of the current version.

The expensive part of starting straight from a template is this: you notice what is missing long after you wrote it, not while writing. The two hardest things to find later are dates and numbers. The first job is an empty note file: the start and end month and year of every role, the employer name, the work you did in that role, the numbers you still have, the tools you used, your education details, your language levels.

While gathering the material you are also making a second decision: which record sits under which heading. That decision changes the result directly, because the class of a record is set by the section heading of the CV. A record under an Internship heading counts as an internship even if its title reads Software Engineer; every record under a Projects heading counts as a project.

The concrete consequence of that classification is the professional duration calculation. Internships, volunteer work, student club roles, education linked roles and projects do not enter it. Part time roles count at half weight, freelance work at full weight. Overlapping ranges are merged, so the same month is never counted twice.

• For every role: employer name, title, start month and year, end month and year.
• For every role: two or three bullets describing the work you did, and a number if you have one.
• The tools and methods you used, under the name you will write on the CV.
• Education: school, department, degree, year, grade average, whether you graduated.
• Languages and their levels; certificates and their dates.

Dates sit in the experience dimension as two separate items: writing the start and end of at least one role as month and year is worth 10 weight, and having a start date on every experience record 6 weight. The second looks at all records, not only the professional ones; a single undated internship or side project record drops that item on its own.

The second job of a date is the duration calculation. Three separate items sit at the 1 year, 2 year and 4 year thresholds of total evidenced professional experience: 10, 10 and 8 weight. A record with no start date, or one whose date cannot be read, never enters that calculation. The record is not ignored, the professional experience item is still earned; duration is the only place it misses.

If no month can be derived from any professional experience, the total duration comes back empty and all three threshold items are lost at once, with the reason that professional duration cannot be computed from the dates. That is how 28 units of the experience dimension budget become invisible. The cause is not a lack of experience, it is a date that cannot be read.

Having a description of what you did under a role is an item worth 8. Having a numeric outcome in at least one experience description is a separate item worth 12, and the heaviest one in the experience dimension. The empty space between a job title and a company name leaves both items unanswered.

What the numeric outcome item looks for is defined: either a percent sign, or a large number with a thousands separator, or a number written with a unit from a fixed list. The list holds: people, users, customers, items, lines, seconds, minutes, hours, days, weeks, months, years, times, points, thousand, million, billion, TL, USD, EUR and their Turkish counterparts. In short, what you are counting has to sit next to the number.

The same sentence can satisfy more than one item. I built the monthly closing sheet in Excel and the closing cycle dropped from five days to three satisfies the description item and the numeric outcome item, and on top of that ties the Excel skill to a usage record. What works is not making the sentence longer, it is putting what you did and what changed inside it.

In the technical dimension, listing a skill is an item worth 6. The same skill appearing in an experience or project record is a separate item worth 14, at least half of the listed skills appearing in usage a third item worth 12, and a skill name appearing in plain narrative text a fourth item worth 14. The last one does not count the technology tag on a record; it looks only at the responsibility, achievement and project description sentences.

Knowing how the search works helps here: the skill name has to appear as it is in the combined experience and project text. The comparison folds Turkish characters into ASCII and lowercases them, and does nothing else. A skill whose folded name is shorter than two characters (the abbreviations R, C, JS) never matches; use the spelled out form.

That is why a difference in spelling between the list and the narrative can drop an item on its own. Writing React 18 in the skill list and React with ... in the narrative means the search cannot find the string react 18 in the text. Write the tool name in the narrative exactly as you wrote it in the list, and leave the version suffix out entirely.

The education section produces not one item but nine: the record (12), the degree name (10), bachelor level or above (14), postgraduate (10), the department name (12), the school name (8), the year (12), the grade average (12), graduation (10). The practical outcome: a line reading A University, 2019 - 2023 satisfies the school and year items and leaves the department and degree items empty. Adding Computer Engineering, BSc to the same line satisfies two more.

On the language side, how you write the level separates two different items. English: good level satisfies the level statement item (18) but not the formal scale item (18). English: C1 satisfies both and brings the third item looking at the C1 threshold (14) as well. The B2 threshold carries 16 more, a second foreign language 10, and a language certificate 10.

Certificates hide one date condition people miss: a certificate dated within the last five years is a separate item worth 14. An undated certificate does not count towards it at all, because a claim of being current can only be evidenced with a date. Write the month and year you earned it next to the certificate name.

• The grade average item only looks at whether it is written, not at its value.
• Programming languages do not go in the Languages section. The extraction pipeline forbids it outright; a Python or SQL written there feeds no language item.
• A language does not have to sit under a heading of its own, but its level does have to be written.

The table below brings together the most frequently broken versions of the rules covered in the five steps.

Do not read the table as a list of penalties. On most rows what is lost is not a mistake but an unanswered question: the engine cannot find the sentence that satisfies the item, so the item stays empty.

Reordering a CV and uploading it again is a common experiment. The result: the same file always gets the same score. The engine is deterministic and 45 items are computed in code. Even on the four items that go to the model, what is asked for is not a score; it is a binary answer on whether it is evidenced, plus a verbatim quote from the CV.

For the score to change, the CV itself has to change: a missing date has to be filled in, an empty description written, a skill sitting in the list tied to the story of a job. How far each of those moves the number was not measured, so there is no gain figure here. The only thing written down is which item looks at what.

When you apply to a company, your CV lands in that company application list. There are two layers there: the fixed evidence list applied to every CV, and a separate fit layer built from the criteria of the posting. The first does not change with the posting, the second comes entirely from the criteria of that posting. The same CV gets the same CV score on different postings and a different fit result.

Whether AI grading is switched on depends on a company setting, so it does not run in every installation. Where it is on, this holds: AI changes the status of no application on its own. It produces suggestions and lists the evidence items and their quotes; the call to advance or reject is made by a person under a user identity.

If you want to see it before sending your own CV anywhere, the free analysis on the candidate side is open and asks for no card. The only accepted format is PDF, the limit is 5 MB and by default 15 pages. There is a daily quota too: 3 analyses from the same IP, 3 for the same user.

• The CV score comes from a list with no position attached; which posting you applied to does not change it.
• Fabricated evidence earns nothing: a quote that does not sit in the CV text is dropped and the item leaves the measurement.
• You can permanently delete your report, your file and your data through the delete my data link on the report page.