AI Content Detection

This guide explains the DrillBit AI Content Detection Report — the downloadable PDF DrillBit generates when a document is checked for AI-generated text. The report estimates how much of the submitted document shows AI-generated patterns, breaks that estimate down block by block, and renders the original document with the flagged sentences highlighted in blue and any manually excluded content in pale cream. Every AI report uses the same four-section structure regardless of the result, and the rest of this guide walks through every parameter you will see.

Throughout this guide the same real report is used as the example: submission 5964872, a 384-word document scoring 33.85 % AI Text Detected, with one of its two blocks manually excluded from the score.

How to Read an AI Content Detection Report

The DrillBit AI Content Detection report is a multi-page PDF. The first page is a one-glance summary — the headline score, where it falls on a 0–100% scale, submission metadata, and the content composition chart. The second page breaks the score down block by block and explains the methodology. Every page after that renders the original document text with the detected sentences highlighted. Here is what the report looks like end to end as you scroll through it:

The report is organised as follows, in this order: a report header repeated on every page; the AI Text Detected headline with its Position on Scale strip; 01 Submission Information; 02 Content Composition; a QR / AI Integrity Help panel; 03 AI Content Analysis — Block Level; 04 How to interpret this report; and finally the Highlighted Document Content. The rest of this guide walks through every parameter in each section.

Report Header & Footer

Every page carries the same header and footer so a printed or partially shared report can always be traced back to its submission. The header shows the DrillBit logo on the left and AI Content Detection Report with the Submission ID on the right; the footer repeats DrillBit • AI Content Detection alongside the same Submission ID. Quote that ID whenever you raise a support ticket — it is the fastest way for the support team to locate the exact report.

AI Text Detected & Position on Scale

The headline block sits at the top of the cover page and is split into two halves. On the left is the AI Text Detected (%) figure — the single most important number on the report. On the right, the Position on Scale strip plots that same figure on a 0–100% range so you can see at a glance how far along the range the document falls. Watch the score count up, then the scale fill in with its value marker:

• AI Text Detected (%): The proportion of the analysed content that shows AI-generated patterns, reported to two decimal places. This is the report's headline result — it is a statistical estimate of AI-pattern signal, not a count of proven AI sentences.
• Matched n words showing AI-pattern signal: The absolute word count behind the percentage. Reading it alongside the percentage tells you the scale of the finding: 33.85% of this 384-word document is 130 words, roughly a long paragraph. The same percentage in a 20,000-word thesis would be nearly 7,000 words and a far more serious finding.
• Position on Scale: A 0–100% strip with the score's position filled in and marked with a value pill. It carries no extra data — it exists so the result stays instantly readable in a printed report or a presentation, without the reader having to mentally place the number on a range.

01 Submission Information

This block records the identity of the document and the person who submitted it. Every field is captured at the moment of upload and cannot be edited afterwards. Watch the animation below as each field is explained:

• Author Name: The name entered for the document's author at the time of submission.
• Title: The document title as supplied during upload. This appears on every downloaded copy of the report.
• Paper / Submission ID: The unique numeric identifier DrillBit assigns to the submission. It is repeated in the header and footer of every page.
• Submitted By: The email address of the account that uploaded the document.
• Submission Date: The timestamp the document was received, recorded in YYYY-MM-DD HH:MM:SS format.
• Document Type: The kind of source DrillBit detected — for example e-Book, Thesis, Assignment, Article, or Synopsis.
• Language: The language profile used to analyse the document. Detection thresholds are calibrated per language, so this field tells you which profile produced the score.
• Text [Pages, Sentences, Words]: Three counts in one field, always in that order — here [2, 43, 384] means 2 pages, 43 sentences, and 384 words of extracted text. The word count is the denominator for everything else on the report: the "Matched n words" line, all three Content Composition percentages, and every per-block figure are all shares of these 384 words.

02 Content Composition

This section splits the analysed document into three mutually exclusive parts and shows them as a donut chart with a matching legend. The note in the corner — Colors match in-document highlighting — is the important detail: the colours here are the same ones used to highlight text later in the report, so the chart and the pages agree visually. Watch the donut draw its three segments and each legend row fill in:

• Human Written — original or non-flagged content (grey, 43.49 %): Everything the detector did not flag. Note the wording carefully: this is content that showed no AI-pattern signal, which is not the same as content proven to be human-written.
• AI Text — highlighted in blue in the document (blue, 33.85 %): The flagged portion. This value always equals the AI Text Detected (%) headline figure, and it is the portion you will see highlighted in blue on the document pages.
• Excluded — manually excluded content (amber, 22.66 %): Content a reviewer or the folder settings removed from the analysis, so it counts towards neither of the other two figures. When this reads 0.00 % nothing was excluded and the score covers the whole extracted document; here it is 22.66%, so nearly a quarter of the document sits outside the calculation.

QR Code & AI Integrity Help

The bottom of the cover page carries a panel with a scannable QR code and an AI Integrity Help section. Scanning the QR code opens the same report on a mobile device, which is useful for showing a result during an in-person review without forwarding the file. Beside it are two permanent reference links: AI Help Centre (drillbitglobal.com/ai-content-detection) for guides, policy and best practices, and Detection Methodology (drillbitglobal.com/ai-method) for the technical explanation of how the score is produced. Both are worth sending to an author alongside the report, so the conversation starts from a shared understanding of what the number means.

03 AI Content Analysis — Block Level

The headline percentage is a single number for the whole document; this section shows how DrillBit arrived at it. The detector divides the extracted text into blocks — contiguous stretches of content — and scores each block separately. Four summary tiles sit above a table with one row per block. In this report one block was included and one was excluded, so the animation also shows how an excluded row is marked:

The four summary tiles read as follows:

• Total Blocks: How many blocks the extracted text was divided into — here 2.
• Included Blocks: How many of those blocks contributed to the AI score — here 1.
• Excluded Blocks: How many were left out of the calculation — here 1. When nothing has been excluded this reads 0 and Included Blocks equals Total Blocks.
• Excluded (%): The share of content the excluded blocks represent, shown in amber. Amber is used consistently for exclusion throughout the report — the same colour marks the Excluded segment in Content Composition, the excluded row in the table below, and the excluded passages in the document pages.

The table below the tiles has one row per block, with these columns:

• Block: The block number. These numbers are not just labels — they reappear as small numbered markers in the document pages, so you can locate exactly where each block starts in the text.
• Sentences and Words: The size of that block — how many sentences and words it contains. These are the flagged and excluded spans only, so they do not add up to the totals in Text [Pages, Sentences, Words]: the unflagged remainder of the document is not a block and gets no row. Here the two rows cover 217 of the document's 384 words; the other 167 are the Human Written portion.
• AI Matched (%): That block's share of the document's total word count, shown as a figure and a proportional bar — block 1's 130 words are 33.85% of 384, and block 2's 87 words are 22.66%.
• Status: Either Included or Excluded, telling you whether the row fed into the headline score. Excluded rows are shaded cream with an amber edge and an amber badge, so they are easy to discount when you read down the table.
• Notes: The line under the table restates the exclusions in words — here 1 blocks (87 words) excluded from the AI score — so the size of what was removed is stated in words, not just as a percentage.

04 How to interpret this report

The final section of the analysis page explains DrillBit's methodology in three short parts. It is worth reading once so you know what the score is measuring — and, just as importantly, why it may differ from another tool's result. The animation walks through each part and the signals behind it:

How detection works. DrillBit combines established AI-detection methods — perplexity analysis (how predictable the next word is, since AI text tends to be smoother than human text), burstiness measurement (how much sentence length and complexity vary, since human writing is more uneven), and stylometric profiling (measurable style fingerprints) — with proprietary techniques developed in-house. On top of those, the engine evaluates four additional analytical layers:

• The hierarchy of writing: how ideas are organised and progress through the document.
• The composition pattern: sentence structure, transitions, and recurring constructions.
• Word prioritisation: the order, weight, and placement of key terms.
• Technical composition: formatting consistency and structural uniformity across the document.

Why scores vary across tools. Detection systems differ in features, training data, and scoring rules, so variation between tools is expected rather than a sign that one is wrong. Certain writing styles — formal, translated, or template-based text — can produce elevated signals in any system, because those styles genuinely share statistical properties with generated text. The example report here is a case in point: a requirement specification made up of short, uniform, numbered clauses is exactly the kind of template-based writing that raises the signal. A raised score on such a document reflects the real statistical profile of the writing, which is precisely why the result needs a human reading rather than an automatic conclusion.

How to use this report. The report is designed as an evidentiary foundation for assessing authorship: it identifies the sections that warrant closer examination and shows the analytical basis for each. Final decisions rest with the reviewer, weighed alongside drafts, prior work, and any other supporting material.

Highlighted Document Content

After the analysis pages, the rest of the report is a faithful render of the original document with the detected content highlighted in the same colours as the Content Composition chart: blue for AI Text and pale cream for manually excluded content. Sentences shown without highlighting are the non-flagged Human Written portion. Small numbered markers indicate where each analysis block begins, so the block table on the previous page maps directly onto the text:

How to use the highlighted content section:

• Match the colours to the chart. Blue is the 33.85% AI Text, cream is the 22.66% Excluded, and everything unhighlighted is the 43.49% Human Written. Reading the pages is therefore a visual check on the cover figures.
• Use the block markers to navigate. The small numbered marker matches a row in the block table, so a block with a high AI Matched (%) can be found in the text immediately rather than by scanning every page.
• Read the highlighted sentences in context. Flagged text often has a noticeably different rhythm, vocabulary, or formality from the surrounding writing. Reading the document straight through, rather than jumping between highlights, usually makes any contrast easier to spot.
• Look at the density, not just the percentage. The visual weight of blue on these pages should agree with the cover figure and with the per-block numbers. If it does not — for instance, a high score but very little visible highlighting — check the Text [Pages, Sentences, Words] field, since a partial text extraction is the usual cause.
• Use the highlights as conversation starters, particularly with student work. Asking the author to talk through a flagged passage in their own words often clarifies whether it was written by them, paraphrased from another source, or generated.