AI Search Visibility

One of the projects the Head of Marketing owns. Not a separate hire.

Owner: Head of Marketing

Accountable to: Yvette Owo, Founder and Managing Director

Works with: Kayla on pipeline, Ash on service detail

Status: Not started. No baseline exists yet.

Also known as: Generative Engine Optimization (GEO), Answer Engine Optimization (AEO), AI search marketing

Where this sits

The Head of Marketing has a portfolio, and this is one line in it. It is the line nobody at the firm currently owns, which is why it is written out in this much detail: the person taking the role should be able to read this and know what good looks like without a second conversation.

The rest of the portfolio (demand generation with Kayla, the website, content, brand, reporting) is written up separately. What follows is this project only.

The project in one line

Own whether SMB Numbers gets named when a small business owner asks ChatGPT, Claude, Gemini, Perplexity, Copilot or Google's AI answers who should handle their books and their taxes.

Why this project exists

A prospect used to type "bookkeeper for small business" into Google and choose from a page of results. Increasingly they ask an assistant instead, and the assistant gives them one answer with three or four firms named in it. If we are not one of those names, we are not in the running, and we never find out. There is no impression, no click, and no line in analytics to tell us we lost.

Kayla fills the pipeline by reaching out to people. This project fills it the other way round, by making us the answer when someone is already asking. The two are meant to meet in the middle.

Nobody at the firm has owned this before, which means whoever picks it up inherits no baseline, no reporting and no bad habits. They build all three.

What you own

1. Measurement, before anything else

Build the prompt set: 50 to 100 questions a real prospect would ask, in their words, not ours. "Who should do my bookkeeping if I run a two person construction company in Texas." "Is it worth paying an accountant to file an S corp election." "Best small business accountant for e-commerce."

Run that set across each assistant on a fixed schedule. Log three things every time: were we named, were we cited with a link, and who was named instead of us. That log is the scoreboard for everything else in this list. We are not interested in activity that does not move it.

2. The machine readable identity

Assistants answer from what they can resolve about an entity, not from what our homepage says about itself. That means structured data on the site (Organization, AccountingService, FAQ, and Person markup for Yvette), the same firm name, address and phone everywhere it appears, a maintained Google Business Profile, and consistent, current listings on LinkedIn, review platforms and the directories that count in our category.

Boring work. It is also the work most firms skip, which is why it still moves the needle.

3. Answer shaped content

Pages built to be quoted. The question as a real person phrases it, the direct answer in the first 50 words, then the specifics that make the answer worth citing: numbers, thresholds, dates, filing deadlines, what it costs, what we would do differently for a $300K business than a $3M one.

Named authorship with real credentials attached. Reviewed and dated, because a model choosing between two sources takes the one it can tell is current. Anything that states a tax or accounting position goes through our internal reviewer before it publishes, the same as any other client facing work.

Depth over volume. Twenty pages that answer a question completely will outperform two hundred that skim.

4. Citations that are not on our own site

Models lean on sources they did not have to take our word for. So: roundups and comparison pages that list firms like ours, review platforms, industry directories, podcast and press mentions, credible Q and A threads where the answer is genuinely useful rather than an advert. Yvette already writes for the Forbes Council, and that surface is under used.

Bring a plan for earning these. Do not bring a list of sites you can pay.

5. Technical access for AI crawlers

Decide, document and maintain our position on GPTBot, ClaudeBot, Google-Extended, PerplexityBot and whoever comes next: who we let in, what they can reach, and why. Keep the site fast, keep the content in the HTML rather than trapped behind JavaScript, and keep sitemaps honest.

You do not have to write the code. You do have to specify the change clearly enough that it can be built and tested, and to check it went live on the site rather than trusting a green build.

6. Classic organic search

Not a separate project. Google's AI answers are still assembled largely from pages that rank, so the traditional work still pays, it just pays through a new door.

The first 90 days

Days 1 to 30. Prompt set built and baselined across every assistant. Full technical and entity audit of the site and our off site listings. One page of findings ranked by impact, with the ten things you would do first and what each is expected to change.

Days 31 to 60. Entity and structured data work done. Listings corrected and consistent. First cluster of answer shaped pages live. Second baseline run, compared against the first.

Days 61 to 90. Second content cluster live. First earned citations off our own site. A repeatable monthly report that Yvette can read in three minutes. An SOP in the hub for every part of this that you have now done twice.

How we will know it is working

The number you own outright: visibility rate, meaning the share of tracked prompts where we are named, per assistant, tracked monthly against the baseline.

The numbers we watch alongside it: citations earned on other people's sites, referral sessions arriving from assistants, and consults booked where the prospect says an assistant sent them. We will add that option to the intake question so the answer stops being a guess.

Attribution in this channel is genuinely immature and we are not going to pretend otherwise. That is exactly why we hold both a number we control and a number the market gives us, and why we look at the trend across a quarter rather than reacting to a single month.

What the person running this needs

• Two to three years in SEO, content or search marketing, with real work you can show us and talk through.
• Direct experience with AI search visibility, or a demonstrable, specific point of view on it. This field is young enough that a sharp practitioner who has been at it for a year beats someone with ten years who is still filing it under "SEO but new."
• Structured data and schema markup, hands on.
• Search Console, analytics, and whatever you use to track assistant visibility. Tell us what you use and why.
• Writing. Most of this job is words, and a page that reads like it was generated will not get cited by the thing that generated it.
• Comfort being measured monthly against a number you set yourself.

Helpful, not required

• Accounting, tax or professional services marketing.
• A working understanding of how retrieval and citation happen inside these systems, so your tactics follow from the mechanism rather than from a conference talk.
• Enough technical range to open a pull request against a Next.js site, or at minimum to write a change specification that a developer can implement without a second conversation.

Not in this project

• Paid advertising.
• Brand and design.
• Outbound prospecting and appointment setting, which is Kayla's.
• Day to day social media management.
• Publishing tax or accounting positions without internal review.

How we work

Fully remote and lean. Everything that gets done twice gets an SOP written for it in our internal hub, so the firm keeps the knowledge whether or not it keeps the person. There is a daily check in and a weekly session on the one thing each of us is making better. You will be expected to bring a number and a next bet to that session, not a list of what you were busy with.

Escalation here means you did the thinking first and now need authority or context you do not have. Bringing a question with no attempt attached is not escalation.

How we will screen

A short practical exercise, paid. We give you a firm, not necessarily ours. You run ten prompts across three assistants, tell us where that firm shows up, where it does not, who is winning those answers instead, and the first three things you would change. One page. We are reading for the reasoning, not the format.

What will end the conversation early

• Guaranteed rankings or guaranteed placement in an assistant's answers. Nobody can sell that.
• No measurement plan, or a plan whose only metric is pages published.
• Bulk generated content as the core of the strategy.
• Paid placement in "AI directories" presented as the strategy rather than as one small tactic.
• Being unable to explain, plainly, why a model cites one source over another.

If this is hired for on its own

• AI Search Visibility Lead
• Generative Engine Optimization (GEO) Specialist
• AI Search and Content Strategist
• Search Visibility Lead, AI and Organic

If the Head of Marketing role is filled by a generalist, this project is the one most likely to be contracted out separately. The first title is clearest internally. The second is what specialists searching for work type into job boards, so it is the one to post under if we want inbound applications.

Something broken

• Tell us what breaks