The draft that could have been written by anyone

You ask an AI tool to draft the company-background section of a proposal. It comes back clean, grammatical, confident. And it opens with a sentence you have read a thousand times: “As a leading provider of end-to-end solutions, we are committed to delivering exceptional value to our clients.”

Nothing in that sentence is wrong. Nothing in it is yours either. Swap your firm's name for a competitor's and it reads exactly the same. The AI produced text that sounds like a company, in general, saying the things companies say. It did not produce text that sounds like your firm, because it was never told who your firm is.

This is the failure mode nobody flags in a demo, because in a demo the output looks polished and the reviewer is looking at grammar, not identity. It surfaces on the third or fourth real document, when a partner reads a draft and says “this is fine, but it doesn't sound like us,” and can't point to a single sentence to fix. The problem isn't a sentence. It's that the AI is writing from nowhere.

Voice is how you write. Identity is who's writing.

These get collapsed into one idea (“brand voice”) and they shouldn't be. They are two different assets that fail in two different ways.

Voice is the surface: tone, formality, sentence cadence, whether you write short and blunt or long and hedged, the phrases your firm reaches for. We've argued elsewhere that voice matters more in AI tools than anywhere else, because generic AI defaults to a bland median register that quietly erases what makes your writing recognisable. If you haven't read it, start with why brand voice matters more in AI document tools than anywhere else.

Identity is the substance underneath: who the firm is, what it does, who it serves, what it believes, the perspective it argues from. Voice tells the AI how to phrase a sentence. Identity tells it who is saying it and on what basis. You can get the voice pitch-perfect and still produce a document that argues from a generic vendor stance, because voice is styling and identity is standpoint.

Here's the tell. A firm with a captured voice but no identity writes fluent, on-brand sentences that make no specific claims. A firm with a captured identity but no voice makes the right claims in a flat, generic register. You need both, and most teams have accidentally captured a bit of the first (by copy-pasting a style note into a prompt) and none of the second.

What an AI-author identity actually encodes

An author identity isn't a paragraph of marketing copy. It's a small, stable set of facts and positions the AI writes from, not about. In practice it covers five things.

1. What the firm actually does

Not the aspirational version. The real one. “A five-person cyber-security consultancy that does ISO 27001 readiness assessments and virtual-CISO retainers for Australian SMEs” is an identity. “A trusted partner in digital transformation” is not. The specific version constrains the AI; the aspirational version gives it permission to write the same fog every competitor writes.

2. Who the firm serves, and who it doesn't

The boundary is as informative as the target. A firm that works exclusively with regulated financial-services clients writes differently, and makes different assumptions about what the reader already knows, than one that serves early-stage startups. An identity that names the audience lets the AI pitch at the right level instead of over-explaining or under-explaining.

3. The perspective the firm argues from

Most firms have a handful of genuinely held positions: the way they think engagements should be run, the thing they refuse to do because it burns clients, the methodology they stand behind. When the AI knows these, a proposal can take a stance instead of staying diplomatically empty. When it doesn't, every document hedges, because hedging is the safe default for a writer with no point of view.

4. The credentials and proof points that are true

This is where identity earns its keep against hallucination. Give the AI the real numbers (years in operation, certifications held, representative client types, the frameworks the team is actually accredited in) and it stops inventing plausible ones. An identity asset is a fence: these claims are true and available, everything else the AI should ask about rather than fabricate.

5. The name discipline

How the firm refers to itself, its offerings, and its methodology. Whether it's “we” or the firm name, whether the framework has a house name that should always be used, whether certain phrases are off-limits. This overlaps with the glossary, and the two work together: the glossary governs terms, the identity governs how the firm presents itself using them.

Where identity shows up in a document

It's tempting to think identity only matters in the “about us” boilerplate. It doesn't. It leaks into every section that makes a claim on the firm's behalf.

The approach section of a proposal reads completely differently depending on whether the AI knows your firm runs a structured discovery phase first or dives straight into delivery. Without an identity, it guesses, and it guesses generic.
The qualifications section is where invented credentials show up. A fixed identity is the difference between “our team of certified experts” (unfalsifiable filler) and “three ISO 27001 lead auditors and a CREST-registered tester” (a claim you can stand behind).
The executive summary is meant to sound like a senior person at the firm speaking directly to a decision-maker. That's pure identity. If the AI doesn't know who's speaking, it writes the summary as a neutral narrator, and neutral narrators don't win work.
Cover letters and closings live or die on whether they sound like a relationship or a template. Identity is what makes “we'd genuinely like to work on this one” land instead of “we look forward to the opportunity to partner with you.”

How to capture it: once, then everywhere

The reason most firms don't have an author identity is that the obvious way to give one to an AI (paste a company description into the prompt every time) doesn't survive contact with a real workflow. It gets shortened, forgotten, or edited inconsistently across team members, and within a month five people are prompting five different versions of who the firm is.

The alternative is to make identity a workspace-level asset, set once and inherited by every draft automatically. In SkyDraft, the author identity (we call it your company info: who the tool is writing as) sits alongside two sibling assets: your brand voice and your glossary. All three anchor every generation. Set them at setup, and every future document inherits them; update one, and every subsequent draft picks up the change. No copy-pasting a style guide into a prompt window, ever.

You don't assemble it from a blank form. The fastest path is extraction: upload one strong document your firm has already written (a proposal you won, a capability statement, an engagement report you're proud of) and the setup pass reads the structure, the brand voice, the glossary, and the identity signals out of it in one go. You review and correct what it inferred rather than authoring it from scratch. The how-it-works walkthrough shows the full setup flow, including how the three workspace assets feed into section-by-section generation.

This is the same principle that underpins the whole workflow: the firm's standard lives in the infrastructure, not in the head of whoever happens to be drafting today. We laid out the full set of pieces a professional-services AI workflow needs in our practical guide to AI document drafting. Identity is the one people most often skip, and the one that most reliably makes a draft sound like a real firm instead of a language model impersonating one.

A quick test before you start

You don't need a tool to find out whether this is your problem. Take your last three AI-assisted drafts. For each one, cover the firm name and the logo, and ask a colleague to read the company-background and approach sections. If they can't tell which of your competitors wrote it, you have an identity gap, not a voice gap, and no amount of tone-tuning will close it.

Then do the opposite test. Read a document a senior partner wrote by hand, one that clearly sounds like your firm. Underline every sentence that only your firm could have written: a specific claim, a stated position, a named method, a real credential. Those underlines are your author identity. The goal of capturing it is simple: get the AI to produce those sentences by default, in every draft, without a partner having to rewrite them back in.

If your firm produces the same shapes of document repeatedly (proposals, reports, compliance evidence, tenders), you already have everything you need to capture an identity. See the use cases we build for for the document types this applies to most directly.