I help businesses in the two places AI is changing fastest. The first is outside your walls: when a prospect asks ChatGPT or Perplexity who handles a problem like theirs, the answer should know who you are and get it right. The second is inside: building the tools your team keeps paying subscriptions for, and getting real work out of the AI licenses you already own.
I came to this from twenty years of doing the work by hand, which is the only reason I know which parts are worth automating.
Not every page that describes your business is yours. Directories, review sites, people-search profiles and old press releases get scraped, copied and repeated. When someone asks an assistant about you, it may lean on whichever version is easiest to read, and if that version is outdated or wrong, you can lose the recommendation without ever knowing it was in play.
I was building websites by hand in the early days of the internet, writing the HTML myself before page builders and plugins did it for anyone. That turns out to be exactly the background this problem needs. Many AI crawlers still read a page the way those early browsers did: the raw HTML the server sends, the structure, and nothing a script adds afterwards. I can look at a site, see what a machine sees, and fix it at the source.
Structured data that tells a model exactly who you are, and one consistent description of your business across your site, LinkedIn and the directories that copy from it. If you share a name with a better-known person or company, this is also where the machines learn to tell you apart.
Finding the directories, profiles and aggregator pages that describe your business, and working out which ones the assistants actually cite. Some you can claim and correct. The rest you outweigh with a clearer, better-sourced version on your own site, built to give a machine good reason to prefer it over the copies.
Whether AI crawlers can reach your pages at all, and whether your content is in the HTML or assembled by JavaScript after the bot has left. Cloudflare now sorts bots into search, training and agents, and choosing Block on the wrong one takes Googlebot out along with the AI crawlers. Worth checking before anything else, because nothing downstream matters if the bots cannot get in.
Assistants pull passages, so a page has to hold up when a single paragraph is lifted out of it. That means the answer near the top, headings that say what follows, specific numbers with a source behind them, and the questions your buyers actually ask answered on the page in plain text. The questions further down this page are an example.
A fixed set of questions your buyers would ask, run across ChatGPT, Perplexity, Gemini and Claude on a schedule, plus AI referral traffic in your analytics and AI crawler visits in your server logs. Answers change from one run to the next, so a single screenshot proves very little. The trend over a couple of months tells you whether the work is landing.
Everything I teach, I run in my own business first. Reporting dashboards, ad account tooling against the Meta API, booking systems, publishing pipelines. Built, in production, doing work I would otherwise pay a subscription for.
It is also what I speak about: a session at Digital Summit Tampa on building software with AI, plus a run of talks on practical AI for marketers at events I host in Tampa, Orlando and Atlanta.
Not a marketing tool, the actual operating system of a small business. Jobs, scheduling, crews, revenue, customer history, all in one place, pulled from whatever software you already pay for and turned into something you can act on. I am building exactly this for a cleaning company right now, because the alternative was four disconnected subscriptions and a spreadsheet.
The dashboard, the report, the data pull that your team currently does by hand every Monday. Most of these are a few hundred lines of code and an API key, and most agencies will sell you a platform subscription instead.
A chatbot that answers from your documentation, your catalog and your policies instead of guessing from whatever it read on the internet. Retrieval-augmented generation, which is a long name for pointing the model at your own material and making it show what it used. Worth it for support deflection, internal knowledge nobody can find, and sales teams who keep asking the same five questions.
Teaching a model to write in your voice rather than the default one. Getting consistent output instead of a coin flip. The reason your chatbot gives mediocre answers is almost never the model.
Practical sessions for marketing teams who have licenses nobody uses well. Delivered to rooms of working marketers rather than to executives who will forward the deck.
Where the security holes are, what should never go into a prompt, and which of these tools should not touch customer data. This is the part of the AI pitch that usually gets skipped, and it is the part that gets companies hurt.
Nobody can promise that, and I would be wary of anyone who does. The models decide what to say. What I can do is remove the reasons they leave you out: pages they cannot read, a description of your business that contradicts itself from one site to the next, and claims with nothing elsewhere on the web to back them up.
A lot of it overlaps. Assistants that search the web lean on the same indexes Google and Bing build, so the technical basics still matter. The newer part is being described correctly. When someone asks an assistant who handles a problem like theirs, the answer depends on whether it knows who you are, what you do, and whether other sites agree.
Blocking training crawlers is a reasonable values call. Blocking the search and agent crawlers opts you out of the answers entirely. On Cloudflare, choosing Block for training now blocks the crawlers that do both jobs, Googlebot and Bingbot included, so check that setting before anything else.
The fixes themselves are days of work. Showing up differently in answers depends on when the assistants next crawl you and how often you are mentioned elsewhere, which takes weeks to months and is not fully in anyone's control.
Ask ChatGPT and Perplexity who you are and what you sell, and send me what they say. Or tell me what your team does by hand every week. Usually one of those things is worth building, one is worth a workflow change, and one should be left alone. Twenty minutes is enough to work out which is which.