A Facebook page is enough for some businesses, and most published advice will not admit it. But it is close to invisible to AI assistants. Facebook allows the crawlers that train AI models and blocks the ones that fetch a page mid answer, so your page cannot be retrieved as a source about you.
Everything written on this question is written by people who sell websites, ourselves included. So treat the usual list of reasons with the scepticism it deserves, and start with the case for saying no.
When a Facebook page genuinely is enough
These are real situations, not hedges.
- Your customers all come from one place and it is not search. A market stall with a local following, a tradesperson who works entirely on referral, a community group. The page is doing the job.
- The buying decision takes under a minute. Nobody researches before buying a coffee. If your category has no comparison step, there is nothing for a website to influence.
- You are testing whether the business works at all. Pre revenue, first six months, still deciding what you sell. Spend the money on finding out if anyone wants it.
- Your whole product is the feed. If the thing people buy is your posting, the platform is the product and a website is a shop window for it at best.
If two or more of those describe you, the rest of this page is not urgent and you can stop reading. Spend the money on something that moves instead.
The argument you have already read
The standard case for a website is control, search visibility and credibility, and it is sound as far as it goes.
| Facebook page | Your own site | |
|---|---|---|
| Cost to start | Free | $60 to $150 a year at the floor |
| Who can remove it | Facebook, without warning or appeal | Nobody, while you pay for hosting |
| Who owns the audience | The platform | You, if you collect it |
| Structure you control | A fixed template | Anything |
| Search visibility | Limited and not in your hands | Yours to work on |
| Reach without paying | Falling for years | Not dependent on a feed |
Consolidated from the arguments in the guides currently ranking for this question, including OnDeck and GetResponse.
All true. Also all a decade old, and none of it is the reason this question has changed.
The part nobody is covering
We read the pages currently ranking for this question. Between them they make the control, credibility and SEO arguments perfectly well. Not one of them mentions AI assistants, and that is now the most important difference between the two options.
When somebody asks an assistant for a business like yours, the answer is assembled from what can be read and verified on the open web. A Facebook page is a poor source for that, for reasons that have nothing to do with quality:
- It is behind a login wall for a lot of content. What a crawler can see of a page is a fraction of what a logged in human sees.
- It has almost no machine readable structure. No schema stating what you do, where you operate, what you charge, or when you are open, in a form a machine reads directly.
- The useful detail lives in posts. Your prices, your service area and your answers to common questions are scattered through a feed in reverse date order, which is the worst possible shape for extraction.
- You cannot fix any of that. On your own site, the nine things that decide machine readability are all under your control. On a platform page, none of them are.
A Facebook page is built to be read by people scrolling. A website can be built to be read by a machine answering a question. Those are different jobs, and the second one is growing.
The reach argument is the one that actually convinces people, and it is the wrong one. Organic reach on a business page has been falling for a decade and everybody already knows it.
The argument I would make instead is narrower. Ask an assistant for a plumber in your town and watch what it cites. It will be directories, review sites, a local paper, somebody’s comparison article. It will very rarely be a Facebook page, and it cannot be yours if you do not have one. That is not a marketing opinion, it is something you can check in about ninety seconds.
What a crawler actually sees on a Facebook page
Rather than assert this, we measured it. On 16 September 2026 we fetched three of the largest brand pages on Facebook with a plain crawler request, no login and no JavaScript, and counted what came back. These are not small businesses, they are the pages with the most resource behind them on the platform, which makes the result more interesting rather than less.
| Page | HTML delivered | Readable words | Schema | Headings |
|---|---|---|---|---|
| facebook.com/nike | 1.19 MB | 1 | none | 0 |
| facebook.com/starbucks | 0.93 MB | 1 | none | 0 |
| facebook.com/facebook | 0.91 MB | 1 | none | 0 |
Our own measurement. Single GET per URL with a generic desktop user agent, scripts and styles stripped, visible text counted. The one word in each case is the brand name in the page title.
Roughly a megabyte of markup each, and one readable word. No JSON-LD stating what the business is, no H1, no section headings, no image markup. Everything a human sees is assembled by JavaScript after login. Our own homepage, fetched the same way on the same day, delivers 2,825 readable words and a schema graph naming the business, the person who runs it, the three offers and their prices, all in the first response.
One detail worth knowing if you ever run a readability tool on a platform page: facebook.com/llms.txt returns a 200 status. It is not a file, it is the site’s HTML shell, which any checker looking only at the status code will score as a pass. We had the same bug in our own checker until we started reading the body.
Facebook allows the training crawlers and blocks the answering ones
This is the part we have not seen written anywhere, and it is the single most useful thing on this page. Facebook’s robots.txt names 35 individual crawlers and then ends with a blanket Disallow: / for everything else. Which AI agents made the list turns out to matter enormously, because the agents do two different jobs.
| Agent | What it does | Allowed |
|---|---|---|
| GPTBot | Trains OpenAI models | Yes |
| ClaudeBot | Trains Anthropic models | Yes |
| Google-Extended | Trains Gemini | Yes |
| Applebot-Extended | Trains Apple Intelligence | Yes |
| OAI-SearchBot | Builds the index ChatGPT search reads | No |
| ChatGPT-User | Fetches a page when a user asks about it | No |
| Claude-User | Fetches a page when a user asks about it | No |
| Perplexity-User | Fetches a page mid answer | No |
| GoogleOther | Feeds Google’s AI surfaces | No |
Read from facebook.com/robots.txt, 16 September 2026. Agents not individually named fall under the final User-agent: * group, which disallows the whole domain. Rules change; check the file yourself before quoting this.
Read that list again in the order it matters. The four crawlers that build training data are allowed. The five that go and fetch a page at the moment somebody asks a question are not. So a model may well have learned something about Facebook pages in general, but when a customer asks an assistant about your business today, the agent sent to look cannot open your page.
There is one real exception and it would be dishonest to skip it. Bingbot is named and allowed, and Copilot draws on the Bing index, so a Facebook page can surface that way. That is a single route, controlled by somebody else, and it is the whole of your coverage.
Facebook is open to the crawlers that train AI models and closed to the crawlers that answer questions in real time. Your page can contribute to what a model knows in general and cannot be retrieved as a source about you specifically.
I expected the robots.txt to be roughly open and it is not, and I would not have found that by reasoning about it. It took four minutes of actually reading the file.
That is the argument I would put to somebody choosing between the two. It is not that Facebook is bad at this. It is that Facebook is making a decision about crawler access that is right for Facebook, and you inherit it. On a domain you own, that decision is two lines in a text file and it is yours. Every one of the nine things our checker looks at is something you set on your own site and something you cannot touch on a platform page.
Check it yourself before you spend anything
This takes two minutes and it is more useful than any article on the subject, including this one.
- Open an assistant logged out. ChatGPT, Perplexity, Copilot, whichever.
- Ask it the question your customer would ask. “Best commercial cleaners in Derby, give me a shortlist with reasons.” Use your trade and your town.
- Read the answer and the citations. Who got named. What sources it leaned on.
- Now ask it about your business by name. See whether it knows you exist and whether what it says is right.
Two useful outcomes. If the answer names three competitors and cites four sources you are not on, you have a concrete list of places to be. If the assistant has never heard of your category being researched this way, you have just saved yourself the cost of a website.
Is a one page site the middle option?
This is not actually a binary. A one page site, with your services, your area, your prices or a range, your hours and a contact form, costs about the annual floor of a domain and hosting and fixes most of the machine readability problem. It will not win you design awards and it does not need to.
Keep the Facebook page. It is doing something a website cannot, which is showing up in a feed where people already are. The two are not competing for the same job.
Common questions
Can a Facebook page replace a website?
For a referral based local business with no comparison step, often yes. For anything researched before buying, no, mostly because a platform page is a poor source for the AI answers and search results that research now runs through.
Do AI assistants read Facebook pages?
Partially and unreliably. Much of the content sits behind a login, there is little machine readable structure, and the useful detail is buried in a reverse chronological feed. None of that is under your control.
Is a one page website enough?
For a lot of small service businesses, yes. Services, area served, prices or a range, hours, contact, and clear structured data. That covers most of what a machine needs to name you.
Will Google rank my Facebook page?
It can appear, particularly for your business name, but you have very little influence over how and no ability to fix the underlying structure. Ranking for what you do, rather than who you are, is much harder from a platform page.
Should I delete my Facebook page if I get a website?
No. They do different jobs. Social reaches people who are already scrolling; a site answers people who are actively looking. Keeping both consistent, especially name, address and phone, matters more than choosing.
If you already have a site
The question is no longer whether to have one, it is whether a machine can read the one you have. Nine checks on your homepage, ten seconds, no email.