Do You Use AI on My Project — and What Happens to My Data?
Exactly what AI is used for here, what it never touches, and what happens to your data. Written to be checked, not to reassure.
Here’s the direct answer
Yes, sosSTEVIE uses AI tools — for research, drafting, code review and analysis — and no, your data is not put into consumer AI tools or used to train anyone’s model. Client credentials, customer records and anything confidential never go into a general-purpose AI service. Every deliverable is reviewed and edited by Stevie before it reaches you, and anything materially AI-generated is disclosed on the item itself.
Key Takeaways
- AI is used here. Pretending otherwise while selling AI services would be an odd way to run this.
- Your data doesn’t go into consumer AI tools, and nothing you send is used to train a model.
- Nothing ships unreviewed. A human reads, edits and takes responsibility for every deliverable.
- AI-generated material is labelled where it’s material to what you’re buying.
- You can opt out of AI being used on your project entirely. It costs more and takes longer, and you’re entitled to choose it.
What AI is used for
Plainly, and specifically:
Research and synthesis. Reading documentation, comparing approaches, summarising long technical material. Anything it tells me gets verified against the primary source before it reaches you, because these tools state wrong things with complete confidence.
Drafting. First drafts of copy, documentation and structured content. Every draft is rewritten. If you’re reading something on this site, a human wrote or rewrote every sentence of it.
Code review and analysis. A second pair of eyes on code, configuration and log files. Useful precisely because it doesn’t share my assumptions.
Pattern-finding in large data. Crawl output, log files, Search Console exports. Thousands of rows, looking for the anomaly. This is the least glamorous use and by some distance the most valuable.
The AI visibility testing itself. When I check whether ChatGPT, Perplexity or Gemini name your business, I’m querying those services. That’s the product, not a shortcut.
What AI is never used for
Your credentials, ever. Passwords, API keys, database contents, hosting logins. These don’t go into any AI tool under any circumstances. They live in a password manager and they’re rotated at handover.
Your customers’ personal data. Customer records, enquiry contents, email lists, anything identifying a third party who never agreed to any of this. If analysis needs it, it’s anonymised or aggregated first, or it doesn’t happen.
Anything under NDA or professional privilege. Several clients are law firms and clinics. Their material doesn’t leave a controlled environment, and where that constrains what I can do, I’ll tell you rather than quietly work around it.
Final judgment. No AI decides what’s wrong with your site, what it’s costing you, or what to do first. Those are the things you’re paying for, and they’re mine to get right or wrong.
Making claims. Nothing on this site — no statistic, no case study number, no capability claim — is published on an AI tool’s say-so. Every statistic here names a source and a year in the visible text, and the reason is that these tools invent citations that look perfect. If it can’t be sourced, it gets cut.
What happens to what you send me
It isn’t used to train anything. Not by me, and the tools used are configured on plans where inputs aren’t used for model training. This is a configuration choice that can be changed by a vendor, so it’s checked periodically rather than assumed once.
It’s minimised. You get asked for read-only access wherever read-only will do. During the Diagnostic nothing on your site is modified at all — that’s a trust position as much as a technical one.
It’s removed. Access is revoked and credentials are rotated at the end of an engagement. If you’d like written confirmation of that, ask and you’ll get it.
It’s kept in one place. Client material lives in a controlled environment, not scattered across chat histories in four different tools.
Disclosure — what gets labelled
Written content: if a substantial passage of a deliverable is AI-generated and only lightly edited, it’s flagged on the item.
Images: any image generated rather than photographed is identified as such on request, and generated images aren’t used to represent real people, real premises, or real results.
Code: AI-assisted code is reviewed line by line before it goes anywhere near your site. It isn’t separately labelled, because reviewed-and-understood code is just code, and the responsibility for it is mine either way.
Analysis and recommendations: never presented as AI output, because they aren’t. If a tool surfaced something, the verification is what you’re paying for.
One rule I hold myself to specifically: no generated image is ever used to depict a real person on this site, and no generated screenshot, chart or metric is ever used as evidence. Every number on the results page comes from a real Search Console export. Fabricating proof on a site whose product is telling you the truth about your website would be self-cancelling.
If you’d rather I didn’t use AI at all
You can ask for that, and you’ll get it.
It means longer timelines and a higher price, because a good deal of the speed advantage disappears — and I’d rather quote that honestly than agree and quietly carry on. The AI visibility testing is the one exception that can’t be opted out of, because querying those services is the deliverable.
Nobody has asked yet. The offer is real anyway.
What I’d ask of you
This runs both ways, and it’s the part most policies skip.
Tell me what your own rules are. If your firm has a policy on AI tools, client data or confidentiality, tell me at the start rather than at review. Several clients have constraints I’d never have guessed.
Check who else already is. Microsoft and LinkedIn’s Work Trend Index found that a large majority of people who use AI at work bring their own tools rather than using something sanctioned — note that carefully, because it’s routinely misquoted as a share of all employees rather than of AI users. Either way the direction is the same: it’s already happening inside most organisations, usually through personal accounts, usually unrecorded.
That’s exactly the problem the Fractional AI Officer work exists to address, and it’s the reason this page is written the way it is. A policy you can’t check isn’t a policy — for your suppliers or for your own team.
How this site treats AI crawlers
AI crawlers are allowed. ChatGPT, Perplexity, Gemini and the rest are welcome to read this site, and there’s an llms.txt at the root written specifically to make that easier — a plain-language summary of who I am, what the four pillars are, the published prices, and the canonical URL for each service.
That’s a deliberate position, not a default. Some publishers block AI crawlers, and for a publisher whose product is the words themselves that’s a defensible call. For a business that wants to be recommended when someone asks an assistant who to call about a hacked website, blocking them is unilateral disarmament.
If you’d like the same for your business — or if you’d deliberately rather not be readable — that’s a real decision with real trade-offs, and it’s part of the AI-readable business work.
Frequently Asked Questions
Is my business data used to train AI models? No. Nothing you send is used to train any model, and the tools in use are on plans configured so that inputs aren’t retained for training. Credentials, customer records and anything confidential don’t go into general-purpose AI tools at all. Vendor terms change, so this is re-checked periodically rather than assumed — if it ever stops being true, this page changes before the practice does.
Does a human actually review the work? Yes, on every deliverable without exception. AI is used for research, first drafts, code review and finding patterns in large data sets; the diagnosis, the recommendations and the final judgment are Stevie’s, and he’s accountable for them being right. Nothing reaches a client unread.
Is the writing on this website AI-generated? Drafted with assistance in places, rewritten by hand throughout. Every statistic on the site names its source and its year in the visible text specifically because AI tools invent citations that look convincing — the sourcing discipline exists as a check against exactly that. If any figure here doesn’t hold up when you check it, that’s a fault worth reporting.
Is it safe to put client information into ChatGPT? Not into a personal consumer account, no. Consumer tiers have historically used conversation data to improve models by default, and for anyone handling client information under a professional obligation — legal, medical, financial — that’s an exposure that exists whether or not anything goes wrong. Business tiers with data processing agreements are a different matter. The distinction is the whole of the answer, and most organisations haven’t drawn it.
Do I need an AI policy for my own business? If anyone on your team uses AI tools for work, then in practice you already have one — it’s just unwritten and set by whoever is using the tools. A written policy needs to cover which tools are approved, what categories of data may never be entered, who decides on new tools, and what gets disclosed to clients. It’s a page or two, not a project.
The short version
I use AI. I’m specific about where. Your data doesn’t go into consumer tools, isn’t used to train anything, and gets removed when we’re done. Nothing ships that a person hasn’t read and taken responsibility for.
If any of that turns out not to be true, I’d want to know before you did.
Questions about this policy: (754) 302-4631, or the contact page.
This page is reviewed quarterly. Last reviewed: August 2026.