Why can a bookkeeper tell at a glance whether Claude got the books right, when most business owners can’t tell whether their SEO agency is doing a good job? Maya Weinreb, founder and CEO of Solvency Now Bookkeeping, answered that on Unscripted Small Business without meaning to. Her team keeps the books for over 150 companies, and she says she can check anything AI gives her because she knows bookkeeping. Then she admitted the reverse: she needs to learn marketing, because otherwise she can’t judge the marketers she hires.
If you sell SEO or content, that’s your client talking. And her rule for AI bookkeeping is the same rule you should be applying to AI marketing: the output depends on the input, and on whether the person giving it understands the work.
Watch the full episode on YouTube or listen on Castos.
Your client can’t check your work the way Maya checks Claude’s
Maya uses AI all day, with agents helping her on all kinds of tasks. She doesn’t use it on client books, for security reasons. But when she looks at AI bookkeeping, she knows what she’s looking at:
“I could look at bookkeeping that Claude churned out and I can tell you if it’s right or not. And Claude does a lot of things right, but it always depends on the input.”
In the same breath she described the other side of that coin. Before we started recording, she’d told me she needs to learn marketing better, “because otherwise when I hire marketing people, I don’t know if they’re doing a good job. I don’t know if what they’re saying makes sense.”
That’s the position most of your clients are in. They can’t audit your keyword research or your content brief any better than they can audit a balance sheet. So the burden of proof sits with you: show the work in terms they can check, like leads, calls and revenue, not a dashboard they have to take on faith. Stephanie Long’s advice on how to hire SEO help that isn’t just vetted by AI is the buyer’s side of the same problem.
Wrong input gets wrong answers, in the books and in your content
Maya’s warning to small business owners who want AI to do their bookkeeping is direct. “If you don’t understand bookkeeping enough, you’re gonna sometimes give it wrong prompts, wrong input, and get wrong answers.”
Her example is worth stealing because it maps so cleanly to marketing. Say you know you made ten thousand dollars and spent eight thousand. The AI can total that. But was the spending on tax-deductible items? Was it expenses on the profit and loss, “or was it balance sheet items that wouldn’t count the same way?” If you don’t know to ask, the tool won’t ask for you.
“The answers are gonna depend on the input and your understanding.”
Swap the nouns and you have the problem with most AI-written content. The model can produce a page on any topic. Whether that page targets the right query, matches the intent and says something new depends on what the person prompting it knows. Andre Savoie made the content version of this argument in AI only works with the known, and if you want to put your own expertise into the input on purpose, build a Claude skill that does your pre-work.
Two AIs gave her two different answers, so check where the data comes from
Maya runs tax research through a specialized tax AI that “pulls data directly from the IRS and state tax laws.” She says she can run the same research in ChatGPT “and not get the same answers.” She trusts the tax AI more. Sometimes she’ll paste its research into ChatGPT and ask it to explain things more simply, but “I’ve had ChatGPT give me wrong answers.”
Her explanation is the part marketers should underline: “it depends where it’s looking for data, it depends on the input you put in, and the only way you’ll know is if you understand it enough to know.”
That’s a practical lens for AI search, too. When a client asks why ChatGPT recommends a competitor, start with the sources the answer pulls from, and whether your client’s information is in them.
“Great aunt Sally Sue” is how buyers see a black-box AI vendor
Maya has heard people say “I’m just gonna have AI do my bookkeeping.” To her, that’s the same as saying “I’m gonna have another bookkeeper or my great aunt Sally Sue or my friend’s neighbor’s cousin’s best friend do it. You don’t know what you’re gonna get. And you don’t know what you don’t know.”
Her complaint about all-AI bookkeeping firms isn’t only accuracy. It’s access: “it’s hard to find someone to talk to when you need to understand your books.” She also told a story she’d heard about an investor-backed AI bookkeeping company that needed outside human bookkeepers to fix its mistakes, until it got so expensive it shut down.
Her pitch for her own firm is the opposite of a black box. “You get replied to within one business day. And if you want to talk to someone, you can talk to someone.” If you run an agency that leans on AI, that’s the bar your clients are holding you to. They’ll accept the automation. They won’t accept not being able to reach a person who understands their account.
Clean records are what make AI useful in your own business
I asked Maya a freelancer’s question. I can download a CSV from my bank, hand it to Claude Code and have it sort my expenses. Isn’t that enough? Her answer: “Claude can do so much now, it’s amazing.” She still recommends keeping a real ledger like QuickBooks Online, because it’s hard to keep everything together otherwise, and because “you need the records so that you can pay as little taxes as possible, but also make sure your taxes are paid on time and paid correctly so you don’t end up in trouble with the IRS.”
Two of her points apply directly if you freelance or run a small agency that pays contractors. These are her views, so check how they apply to you with your CPA:
- Get a W-9 first. She says the 1099 threshold just changed from $600 to $2,000, indexed to inflation, and her rule is to “make sure you get them to fill out a W9 before you pay them so you can send them a 1099 in January.”
- Let your records settle the mismatch. When I asked what happens if a payment processor reports more than your 1099s show, her answer was “first you have to have really good records,” so every dollar reconciles to your bank accounts.
Treat the side project that earns like a company
Near the end I asked Maya about rental property. My father-in-law may be moving closer, and I’d just interviewed my friend Alex Zweydoff of ClearLead Digital, a property management marketing agency, about how owners pick a manager. (His take for marketers is in owner intent over tenant traffic.) Should rent come in through an LLC or as regular income?
Maya started by clearing up a common mix-up: “There’s literally no tax benefits to doing an LLC, but there’s legal benefits to having an LLC because it’s a limited liability company.” Then she widened it: “if you have anything that’s making you money, you should treat it as a company.” That applies to the affiliate site, the template shop or the consulting you do on the side. If it earns, it needs its own records, so you can decide whether to invest in it, sell it or shut it down.
Questions to ask before you hand AI the job
- Do you understand the work well enough to tell whether the output is right?
- What input are you giving it, and what does that input leave out?
- Where is the tool pulling its data from, and would a specialist source give a different answer?
- When the client has a question, who do they talk to, and how fast do they hear back?
- Can your client check your results in their own terms, or are they taking them on faith?
The full episode recap is Maya Weinreb on AI bookkeeping, clean books, 1099s and taxes on Unscripted Small Business, and my own take is on knowing enough to check the work. You can find Maya’s firm at Solvency Now Bookkeeping, including its bookkeeping services for small businesses.
This episode is supported by our sponsors: uSERP, a GEO agency that helps brands get found in AI search, and Oleno, which gives a whole team the same shared context so everyone uses AI the same way.
Sponsor disclosure: uSERP, Oleno and ClearLead Digital are paid sponsors of the Unscripted podcast network. Links to them are marked as sponsored.