This is Part 1 of a three-part series on what actually happened when I started using Claude, Anthropic’s AI assistant, as an office assistant for my one-person photography and video business. Part 2 covers the infrastructure behind it, the backups, the second brain, and the personalised skills, and Part 3 is the honest part: the pitfalls, the questions I get asked, and where I take this next.
I run Greg Harding Photography + Video on my own. No studio manager, no account handler, nobody else picking up the phone. Every enquiry, quote, invoice chase and follow-up email lands on my desk right alongside the actual photography and video work, and for years the admin side quietly ate into my shoot days. A couple of hours here, an evening there, and a week that should have had three shoot days and two office days had somehow flipped.
I started at the beginning of July, so this is just over two months in, and trialling Claude properly was only possible because of a quiet spell over the summer holidays, when bookings slow down and I finally had the time to build something instead of squeezing it in between shoots. I'd already had a look round the wider field, ChatGPT, Copilot, Gemini, Perplexity, and checked what each one's usage settings actually let me do before I sank any real time into one. I never wanted it replacing my judgement on client work, or publishing copy I hadn't read. What I wanted was help with the repetitive scaffolding: quoting, keeping the CRM tidy, first drafts of follow-ups, and the website and SEO admin I'd been putting off for months. I built all of it myself, no IT background, no coding specialist, nobody brought in to set it up for me. Where I got stuck, I asked Claude itself to refine a prompt or review a skill rather than muddling through alone, which turned out to matter more than any technical knowledge I didn't have. This is the first of three posts on what actually happened over those two months, the good bits and the properly annoying ones, written for other creatives doing the job of three people who are wondering if an AI office assistant is worth the setup faff.
I aim for three shoot days and two office days a week, though the mix shifts depending on distance between jobs and post-production time, which ranges from half a day for a single stills shoot to a day and a half for a two-day one. Admin, invoicing and chasing people on the phone never got their own block. They were the hour or two squeezed in around everything else, and the busier the diary got, the more that squeeze built up. That's the problem I was actually trying to solve.
Honestly? Roughly broke even so far. Weigh the hours it's freed up against the hours spent building, testing and fixing what's behind it, and it's close enough that I won't call it a clear win yet. Not the answer this kind of post is meant to end with, but I'd rather say it straight than talk it up. What gives me confidence it'll tip the right way: the setup cost was one-off, the weekly saving isn't, and the single most useful thing to come out of it has already outlasted several rounds of fixes and rebuilds, more on that in Part 2.
The very first thing it did, on day one, wasn't build anything, it diagnosed. It went through what I already had and found I was sitting on a stack of paid-for assets I wasn't actually using: a sector portfolio that had never gone out in a single sales response, published work never once cited in a pitch, real client reviews sitting there unused with not one quote pulled into an email, years of finished films with no way for a prospect to see the right ones for their sector. The instruction that came back was blunt, use what you already have before producing anything new. Not a flattering thing to admit about your own business, but accurate, and it reset how I thought about everything I built after it. Most of what's below exists to put assets I already had to work, not to generate more of them.
The best thing I've built turns one shoot brief into a handful of finished outputs. After a job I run what I call a Wrap Sheet, a five-part prompt covering the brief, the project's story, the images, any video, and a quick summary of how it went. That becomes a LinkedIn post, Instagram captions, a case study blog post and a client follow-up email, all drafted and waiting for me to read, edit and approve, plus a nudge to ask for a Google Business Profile review while the job's still fresh rather than three weeks later when it's gone cold.
New enquiries get a similar treatment: a bit of research on the company, roughly what sector and scale, and a first read on whether it's worth chasing hard or handling at arm's length, before I've even opened the email properly.
The CRM is the second win, and I built this one from scratch with Claude rather than adapting something off the shelf. Enquiries and jobs run through Gmail and a Google Sheet, the "GHP+V Management Dashboard," with tabs for clients, jobs, live enquiries, archived enquiries and completed work. Claude drafts quotes, updates rows and keeps notes current so I'm not typing into every cell myself. A signed booking agreement goes out before I hold a date for anyone, not just a verbal yes, and a deposit gets requested automatically once a job crosses a certain size or involves travel, so the decision about who pays what and when isn't something I'm working out fresh for every new client. Enquiry and aftercare follow-ups (day 0, 3 and 10 nudges) fire off on their own. A check every couple of hours through the week keeps it all in sync, so I'm not the one who has to notice when something's slipped.
Every morning I also get the Life Dashboard, a short briefing pulling together things like where cash sits via Xero and flagging anything that looks obviously off before I've had my coffee. Not glamorous, just useful, the way a good PA flags something on their way past your desk.
The SEO and website work has been the slowest burn but the easiest to point a number at: audits, Search Console and GA4, reviving dormant blog posts, building proper pages, and going through keyword titling and alt text on images, both in a worse state than I'd realised until someone actually checked. Organic sessions have grown, and more interestingly, so has referral traffic from the AI tools themselves, ChatGPT, Gemini, Perplexity and Claude are all sending people to my site now. For a one-man band up against agencies with a full marketing team, that's not nothing. I've also got a routine that posts to my Google Business Profile on a schedule now, rather than it being the thing I always meant to keep on top of.
A few jobs it's handled better than I expected, and better than I'd have managed myself in the same time. One project needed a RAMS, a risk assessment and method statement, for several different sites, each with its own access, hazards and rules. Rather than copying one document and swapping the address, it built each one properly, tailored to that site, formatted correctly. Normally a slow, fiddly job I've put off more than once. Not this time.
Anything involving tables, numbers or checking something against an online reference, it's strong at: keeping a table consistent across several sites, pulling the right detail from a source rather than guessing.
It also fixed how I send clients my terms before starting work. I already had a full set of Terms and Conditions a solicitor had drafted years ago, multi-page, properly worded, and barely sent because it made clients nervous before we'd even started. Claude took that same document and turned it into something short and plain that still covers what needs covering, without touching what it actually meant. Signing a contract stopped being the awkward bit of onboarding a new client.
Part 2 covers the less glamorous half of this: hardening the backups on my Synology NAS, building a client delivery system that finally gets me off WeTransfer, and the "second brain" wiki that's turned out to be the single most useful thing I've built in these two months, plus what it's actually like day to day having that in place. Part 3 is the honest one: what's actually gone wrong, the questions people keep asking me about this, and where I take it from here.