Short answer: ChatGPT recommends the fractional CFOs whose pages it finds and can quote when it searches the web. Publish one page per question your clients ask, answer it in the first two sentences, add dated facts it can attribute, keep the same description on the profiles and directories it reads, and let OpenAI's search crawler access your site. Then measure every month, on five real client questions, and judge after three months.
Founders now ask ChatGPT the question they used to ask their accountant or their investor: "Who could be our part-time CFO? We're a 25-person SaaS company in Lyon on Pennylane, raising a Series A next year." What comes back is a short list, a few names or firms, and the sources ChatGPT read to build it. This guide is about being on that list.
We are not a marketing agency. We are a software company that has measured, every month since August 2026, which sources ChatGPT, Claude and Gemini cite on a fixed set of finance questions. What follows is what those measurements show, applied to a fractional CFO practice.
How Does ChatGPT Decide Which Fractional CFO to Recommend?
It answers from one of two places, and only one of them is in your hands this year.
From what it learned in training. When ChatGPT answers without searching, it draws on what the model absorbed before its training cut-off. Firms that were written about a lot, for a long time, come up. For an independent practice, this layer moves slowly: what you publish today may reach it in a later model, months or years away.
From the pages it reads when it searches. For a request with a place, a sector or a recent date in it, ChatGPT usually searches the web, reads a handful of pages and cites them next to the answer. This layer moves in weeks. If your page answers the question well and the engine can reach it, you can be among the sources quickly.
A detail from our own measurement shows how much the two differ. When we first ran our question set through OpenAI's API with web search merely allowed, the model answered most questions from memory: long, plausible answers, with zero searches and zero sources. Once search was required, the same question produced six searches and four citations. The recommendation a founder sees depends on which of the two modes ChatGPT chose for their question. Specific questions push it toward searching, and that is good news for a specialist practice.
OpenAI runs a dedicated crawler for this, OAI-SearchBot, and states that sites which block it do not appear in ChatGPT search answers. In practice, the old basics of being findable on the web still apply. They are the entry ticket, not the whole game.
What Our Citation Tracking Shows
Every first Monday of the month, a script asks the same 15 questions to ChatGPT, Claude and Gemini with web search on, and records every source each one cites. Four runs so far (August, September, October and November 2026), 45 answers per run. Three patterns are relevant to a fractional CFO.
1. Each answer cites a handful of sources. ChatGPT cited 5.2 sources per answer on average, Claude 4.4 and Gemini 4.3. Across four runs, 452 different domains were cited. The list is short for each answer and long overall: there is room, but each slot is contested.
2. The list changes every month. Of the sources cited for a given question by a given engine, only 18 to 33 % were cited again the following month. One snapshot tells you very little. A practice that checks once, sees itself absent and gives up has measured noise.
3. Owning a definition works. Our site was cited in 6 to 9 of the 45 answers each month. On ChatGPT, one question cited us in all four runs: "What is FINANCE.md in financial modeling?". FINANCE.md is a convention we named and documented. Nobody else answers that question better, because nobody else defined it. On generic questions, we compete with much larger sites and mostly lose.
A caveat we keep in front of every reading: the API search we measure is close to, but not identical to, what a logged-in user sees on chatgpt.com. Treat the patterns as directions, not as a ranking algorithm.
Eight Steps to Get Your Practice Recommended
1. Write down the questions your best clients asked before hiring you
Not keywords: questions, in their words. "We're about to raise and our numbers live in three spreadsheets, who can fix that?" "Do I need a CFO or a better bookkeeper at 2 million in revenue?" "Who can run our month-end close on Xero for a few days a month?" Five is enough to start. These are the prompts founders type into ChatGPT, and the rest of the work hangs on them.
2. Make your niche explicit
ChatGPT recommends against a specific request. "Fractional CFO, FP&A, fundraising, cash" matches every request a little and none exactly. Sector, stage, country or city, the ledger you work on, the problem you solve: say them plainly, on the page and in your profiles. A practice described as "fractional CFO for venture-backed SaaS companies in the UK, from seed to Series B, on Xero" gives the engine something to match.
3. Publish one page per question, answered in the first two sentences
The pages that get cited answer the exact question near the top, then explain. A services page answers no question in particular. A page titled "When does a SaaS company need a fractional CFO rather than a bookkeeper?" answers one, and can be quoted in a sentence. Put the answer first, the nuance after, and use the question itself as a heading.
4. Use dated, specific facts
Engines cite what they can attribute: a number with its date and source, a threshold, a procedure with steps. "We help founders make better decisions" is never cited. A sentence such as "For the companies we support, the first board pack takes three weeks to set up and two days a month after that (2026, 11 clients)" can be. Anonymise client figures and get consent before publishing anything derived from a client's numbers.
5. Name something and own its definition
This is the lesson from our steadiest citation. A named method, a checklist, a template, a framework your clients already hear you describe: give it a name, write the definitive page about it, and keep that page current. "The 90-day cash review" or "the board pack in five pages" is more citable than a fifth article on "why cash is king", because you are the reference for it.
6. Be present where ChatGPT looks
In our runs, directories, marketplaces, code repositories, public profiles and association pages come up again and again among the sources. For a fractional CFO, that means the platforms where clients compare providers, a complete LinkedIn profile, the member directories of the networks and professional bodies you belong to, and podcasts or guest articles that link back to you. Use the same name, the same niche sentence and the same location everywhere. An engine that finds three consistent descriptions trusts the entity more than one that finds three different ones.
7. Fix the technical basics once
- Let OpenAI's search crawler in. OpenAI separates OAI-SearchBot (used to show sites in ChatGPT search) from GPTBot (used for training). Blocking the first in
robots.txtkeeps you out of search answers. You can block training and still allow search. - Be indexed by the major search engines, including Bing: register your site in Google Search Console and Bing Webmaster Tools.
- Publish in plain HTML. A case study inside a PDF or an image is harder to read and quote than the same text on a page.
- Describe the practice in structured data (a
ProfessionalServiceorPersonschema with your niche, location and profiles), so the facts about you are unambiguous.
8. Measure monthly, judge after three months
Ask ChatGPT your five client questions on the first working day of each month, from a logged-out window if you can, and note who is cited. Keep the questions fixed: changing them breaks the comparison. Given how much the lists move from one month to the next, judge the trend after three months, not after one.
What Does Not Work
- A single services page. It answers no question precisely, so it is rarely the best source for any of them.
- Stuffing pages with "AI" and "ChatGPT". Engines quote answers, not keywords. A page about ChatGPT is not a page ChatGPT recommends.
- Generic posts on generic topics. "Five tips to manage cash flow" competes with thousands of larger sites. A precise answer for your niche competes with a handful.
- Judging on one search. The cited sources change from week to week. A single check, good or bad, is an anecdote.
- Paying for "AI ranking" guarantees. Nobody controls what an engine cites. What you control is whether your page is the best answer to a precise question, and whether the engine can read it.
What Should a Fractional CFO Publish First?
If you have a week, not a quarter, start with these three pages:
- Who you help, in one sentence, on a page of its own. Sector, stage, location, ledger, problem. This is the page an engine matches against "fractional CFO for [your niche]".
- The answer to the question you hear most in discovery calls. Often "Do we need a CFO yet?" or "What does a fractional CFO actually do each month?". Answer it for your niche, with one dated example.
- One named method, documented. The thing you do at every client, written as a procedure with steps. It is the page most likely to become the reference.
A worked example helps all three: an anonymised model, a template a founder can open, a sample board pack. A visitor sees how you work, and an engine finds concrete content to cite. If you need one, we publish a fractional CFO financial model template you can adapt under your own name.
Why This Matters Now
Your clients and your peers already live in these tools. In the 2026 Fractional CFO Study (173 practitioners, 40 countries), 60 % of the 140 fractional CFOs who named their tools use ChatGPT, and 79 % of 152 respondents used an AI assistant on client work every working day. The founders who hire them are on the same tools. The question they type is the new first meeting, and the sources ChatGPT cites are the shortlist.
Where Layerz Fits
Layerz is a structured spreadsheet, ready for AI, and an official ChatGPT plugin. It does not get your practice recommended. It helps with what a recommendation leads to: each client's model persists between sessions, ChatGPT or Claude can read and update it, every change is in the version history, and the model exports to clean Excel with formulas intact. A model you can share as a read-only link is also a concrete way to show prospects how you work.
How fractional CFOs use ChatGPT on client work, rather than to be found, is covered in ChatGPT for Fractional CFOs.
It is free for one person working alone, with unlimited models.
FAQ
How do I get my fractional CFO practice recommended by ChatGPT? Publish one page per question your clients ask, answer it in the first two sentences with dated, specific facts, make your niche explicit, keep the same description on the profiles and directories ChatGPT reads, and allow OpenAI's search crawler on your site. Then check every month who ChatGPT cites on five real client questions.
How does ChatGPT choose which businesses to recommend? Either from its training data, which changes slowly, or from the pages it reads when it searches the web, which it cites next to the answer. Specific requests (a sector, a city, a ledger) make it more likely to search, so pages that answer specific questions have the best chance.
How long does it take to be recommended by ChatGPT? For search-based answers, weeks after a page is published and indexed. Because the cited sources change a lot from month to month, judge after three months of monthly checks. For answers from training data, it can take a model generation or more.
Should I block ChatGPT from crawling my site? Not if you want to be recommended. OpenAI uses OAI-SearchBot for search results and GPTBot for training; you can block GPTBot and still allow OAI-SearchBot.
Is getting recommended by ChatGPT the same as SEO? Mostly the same discipline, with a different reader. A search engine ranks your page; ChatGPT quotes a sentence from it. The sentence that answers the question, near the top, matters more than the keyword.
Do directories and LinkedIn matter for ChatGPT recommendations? Yes. Directories, public profiles and association pages are frequent sources in AI answers. Keep the same name, niche and location on all of them.
The Takeaway
ChatGPT does not pick fractional CFOs at random. It cites the pages that answer the founder's question precisely and that it can reach. Be specific about who you help, answer real client questions in the first lines, name and document the method you are known for, keep your profiles consistent, and measure every month. The practice that does this steadily for three months will usually see more of its own pages among the sources, and every one of those pages also works for the founder who reads it.