Articfly

AI articles hallucinate. 3 sources fixed mine

Across 63,000 articles shipped on my stack, the fastest way to lose a client was one made-up statistic in a published post.

People worry whether AI content sounds human. Wrong worry. The real risk is whether it is TRUE. A confident article that cites a fake study or a number the model invented does more damage than one that reads a little robotic. Google's EEAT guidelines punish it, and a client who catches one hallucination stops trusting the whole pipeline.

So the credibility problem is not a writing problem. It is a research problem.

Here is what actually changed the output.

Every article gets grounded before a single sentence is written. The pipeline runs live research first: Brave Search for current facts, Wikipedia for stable entities, Perplexity for synthesis. Real sources, pulled in, not recalled from training data.

The model then writes against the retrieved facts, not from memory. If a claim is not in the research context, it does not make it into the draft. That one rule kills most hallucinations.

Entities, dates, and stats come from the source set. The model fills sentences, it does not invent the numbers inside them.

The result across 9 retainer clients: articles that rank, get cited in AI search answers, and survive a fact-check from a client who knows their own industry. That last one is the test most AI tools fail.

This is the part I built articfly.com around. Grounded first, written second.

If you are shipping AI content at volume, what is your actual process for catching a hallucination before it goes live, or are you just trusting the model and hoping?

Hey everyone!

When it comes to payments, almost everyone defaults to Stripe. I didn't.

Don't get me wrong—I’m not saying Stripe is bad or too expensive. It’s a great product. But I decided to go with Polar.sh instead, and it came down to one absolute dealbreaker for me: Merchant of Record (MoR).

Here is why that won me over definitively:

• Total Tax Peace of Mind: As an MoR, Polar acts as a reseller. This means they issue the invoices to the customers and handle all the VAT, sales tax, and other annoying bureaucratic nightmares.

• Convenience Over Everything: I know Stripe has "Stripe Tax," but MoR is a step further. It takes the liability completely off my plate. Every indie hacker knows that when you finally start generating revenue, the absolute last thing you want is a stressful audit or a meeting with the tax office. I just want to write code and build my product.

• Discount Codes on the Free Plan: This is the second biggest reason. Polar includes discount codes right out of the box on their free tier, which is incredibly helpful when you are just starting out and trying to get those first early-bird users.

Has anyone else made the switch to an MoR like Polar or LemonSqueezy? Curious to hear your thoughts on this vs. standard Stripe!

Hey Indie Hackers,

I recently reached out to my clients to get some feedback on what features I should add and what I could improve.

A few of them replied, and the biggest takeaway was really surprising: they told me the app itself is great, but my value proposition tells them absolutely nothing. I really tried to make my initial copy as honest and straightforward as possible, but clearly, it missed the mark!

So, I took their feedback to heart and rewrote my pitch. Here is the updated version:

Articfly: Your blog on autopilot. Get a full month of high-converting content that attracts new leads every day. Just connect to Articfly, and we handle the rest. We analyze, plan, and publish blog articles tailored specifically to your needs so you can hit the first page of search results. No long-term commitments—you can cancel anytime.

What you thinks?

I shipped 50,000 AI-written articles through my tool in the last 12 months. So when people post the dead internet theory on this forum, I am the villain in the story.

After a year of watching how customers actually use the thing, my honest take: half of it is true. The other half is paranoia.

WHERE THE THEORY IS RIGHT

Content farms got worse. Before GPT, a farm needed five cheap writers on Upwork. Now it needs one prompt and a CMS. Google page one for low-value queries is mostly garbage. Comment sections on big sites are 30 to 50 percent bots talking to bots. Reddit is laundering generated content as personal stories. LinkedIn is a hallucinated thought-leader loop.

I will not pretend I have nothing to do with this. My tool, in the hands of someone who clicks generate, picks the cheapest title, and publishes without reading, makes the problem worse.

WHERE THE THEORY IS WRONG

The panic acts like every AI-written word is identical garbage. It is not. From my logs:

- 70 percent of articles get manual edits before publish. People rewrite intros, fix specific claims, swap headlines.

- The top 20 percent of accounts edit harder than they generate. They use the tool for a 60 percent draft and spend an hour on the post.

- The 10 percent who publish raw also see the worst traffic. Google demoted that layer years ago.

The internet that pays (places where someone actually buys something, signs up, or hires you) does not run on raw AI slop. It runs on AI as a draft layer plus a human who knows what they are saying.

THE LINE NOBODY DRAWS

There is a difference between AI content and AI-assisted content. The dead internet panic blurs them because the panic sells better as one big enemy.

Raw GPT output ranking for "best CRM 2025" is dead-internet content. The same person using AI to skip the empty first draft and then spending real time on the post is just modern writing.

My nine retainer clients running AI-assisted content see 12 to 28 percent organic traffic lift year over year. Pure AI-slop sites are getting hit by every Google update since March. Both things are true.

WHAT I THINK ACTUALLY HAPPENED

The internet did not die. The cheap layer of it just got cheaper. Top-of-funnel SEO content that was always low value is now even lower value and produced 100 times faster. That feels like death because the volume is overwhelming.

The parts that actually matter (specific expertise, real numbers, named people, brand voice) are harder to fake and still rare. That is where humans still win. That is what I tell customers to focus on. Articfly is the draft tool, not the publishing tool.

So yes, the internet got more polluted. No, your blog is not pointless. Write things only you can write, generate the parts where the draft is the bottleneck, and stop pretending AI is either the enemy or the savior.

What is your read on the dead internet theory? 30 percent true, 50, 80?

Can AI articles not get flagged as AI?

Short answer: no. Long answer: yes.

Most articles, whether written fully by AI or by a human, are built differently. The structure gives them away. Most AI content tools market themselves as "anti AI detection" or "humanize your AI". That claim is not true. I have tested all of it.

AI writes very well. The output is good. But it still leaves artifacts. Sentence rhythm. Connector words. Paragraph shape. Vocabulary range. The architecture underneath is detectable no matter what you do on top.

You can fine-tune the model. The artifacts stay. You can run a humanizer pass. The artifacts stay. You can chain three LLMs through three rewrites. The artifacts stay. The score on a detector moves a few points. The truth does not change.

So why keep pretending?

The competitors selling "undetectable AI" are running the same OpenAI and Anthropic models everyone else has. Same backend. Different paint job. The "undetectable" label is marketing, not engineering.

Here is the other path.

Stop using the AI as the writer. Use the AI as the assistant. The article should sound like a human wrote it with AI help, not like AI wrote it pretending to be human. That is what I did.

What changes when you flip that role:

1. A human brief sets the structure section by section, not a single prompt.

2. Each section gets its own context. Audience. Angle. Word count. Required entities. Banned phrases.

3. The model fills the section against that brief. No free monologue.

4. A brand voice profile, extracted from real human writing the user has already published, guides every paragraph.

5. A final pass strips the obvious AI tells.

The model still does most of the typing. The steering is human all the way down.

The result: SEO did not suffer. AEO did not suffer. The articles work. They rank. They are visible in AI search surfaces. That is the only test that matters.

If you are building in this space, stop selling undetectable. Start shipping articles that read like a human wrote them and rank like a human wrote them. The classifier game is a dead end. The structure game is the real one.

I built articfly.com around this. AI-assisted, never AI-authored at the wheel.

What do you think happens to the "humanizer" category in the next 12 months? Does it die when the obvious tells get cleaned up at the model level, or does it keep selling on hope?

One years ago I sold a $29 n8n template on Gumroad. It outsold the SaaS I had spent six months building. That single Gumroad listing turned into a three-brand stack I now run solo from Poland.

Here is the actual sequence, with the numbers I can verify.

THE FAILED PROJECT

I had shipped a B2B tool. Forget the name. It got 11 signups. Two paid. Zero retention. I was burned out and broke.

On the side, for my own sanity, I had built an n8n workflow that researched a topic, drafted a 1500 word blog post, and pushed it to my WordPress draft queue. It was glue code, not a product. I wrote the README in 40 minutes and put it on Gumroad for $29 because I needed coffee money.

THE $29 SALE

It sold 14 copies in the first week. The dead SaaS sold zero in the same week.

The interesting part was the email after the sale. Three different buyers wrote some version of: "I cannot get this running. Will you set it up for me for $500?"

I said yes to all three. That is when I learned the template was not the product. The setup was the product. The template was the lead magnet.

THE SERVICE SPIRAL

Over the next four months I took 9 setup gigs at increasing prices. $500 to $2,800 each. Every client wanted small variations. Different research sources. Different brand tone. Different CMS targets.

I started naming this work Lumizone. One years later Lumizone has shipped 1,000+ articles across 9 active retainer clients. One pest control client got +15% organic traffic in a 6 month window. One peptides e-commerce client got +19%. These are clients in industries nobody wants to write about, which is exactly why the SEO is winnable.

THE 80 PERCENT MOMENT

After the fifth custom build I noticed something. The architecture was the same on every project. Research stage. Outline stage. Per-section briefing. Draft. Score. Publish. Refresh. The only things changing were the brand voice profile and the CMS target.

I had built the same pipeline 5 times.

So I extracted it. Same architecture, no per-client work, one dashboard, one price. I called it Articfly. The first 50 users came from existing Lumizone clients who wanted the cheaper self-serve tier. articfly.com is the product I would have built first if I had known what to build. 50,000+ articles have shipped through it since.

WHAT THE GUMROAD SALE ACTUALLY TAUGHT ME

3 lessons, in the order I learned them. None of them are what I expected.

1. The market tells you which problem is real. You do not get to vote. My polished SaaS got 11 signups. My 40 minute Gumroad listing got buyers asking to pay 100x for setup. I had been working on the wrong thing for 6 months.

2. Productize the bottleneck, not the tool. The tool was easy to copy. The bottleneck was setting it up correctly for a specific brand. That was Lumizone for 2 years. The SaaS version only worked once I had productized the bottleneck itself: brand voice analyzer, automatic CMS adapters, per-section briefs.

3. Service to product is a one way street, but it takes longer than you think. I should have started extracting Articfly after client 3. I waited until client 5. That cost me roughly 8 months of compounding.

The current shape: Lumizone for custom work where the client wants a hand on the wheel, Articfly for everyone else who wants a product they can run themselves. Same engine underneath. Different price points. Different buyers.

I would have laughed in 2024 if you told me a Gumroad listing I wrote during a bad week would split into two businesses.

Question for anyone who has done this loop: at what point did you stop taking service work? I am at the point where Lumizone retainer cash is high but Articfly compounding is real. Pulling the plug on the agency feels both obvious and terrifying.

AI search killed SEO. My 9 clients disagree.

Last quarter my pest control client got +15% organic traffic. My peptides e-commerce client got +19%. Both rank on Google AND get cited inside ChatGPT answers.

Every other week someone on this forum says SEO is dead and you should just optimize for AI search instead.

I run a content agency and a content SaaS. I see traffic data for 9 active retainer clients and 50,000+ articles shipped through production. Here is what I actually see.

WHAT IS DYING

Top-of-funnel definition queries. "What is X." Google's AI Overviews now answer those without a click. If your blog ranked for "what is content marketing" last year, that traffic is gone.

Generic listicles. "10 best tools for X." AI summarizes the list before anyone scrolls.

WHAT IS NOT DYING

Bottom-of-funnel buying queries. "Best X for Y industry under $Z." People still click 3 to 5 results because the decision matters and they want to compare.

Niche long-tail with specific intent. "How to fix WordPress 502 after a Cloudflare update." Nobody wants a 4 line AI summary for a 45 minute fix.

Brand and comparison queries. "X vs Y" still drives signups even when AI shows a preview. People want to verify.

WHAT ACTUALLY CHANGED

The job is no longer rank on page 1. The job is be the source AI tools pull from.

ChatGPT cites Reddit, product pages, Quora, and authoritative blogs. Perplexity cites the same plus trade press. Google AI Overviews cite top 5 organic results. If your content does not exist, you cannot be cited. If it ranks on Google, you get cited everywhere else for free.

THE NUMBERS THAT MATTER

50,000+ articles shipped through articfly.com. Across 9 retainer clients, average organic traffic is up between 12% and 28% in the last 6 months. Not because we gamed anything new. Because we kept publishing while half the market froze and waited for clarity.

The agencies that paused SEO budgets last year are losing pipeline right now. The ones still shipping are eating it.

MY OPINION

SEO is not worth it in 2026 if you write for clicks on definition queries. SEO is worth it in 2026 if you write to be referenced by both Google and AI engines. Different game. Same channel. Same content infrastructure.

Question for anyone who has cut SEO budget this year: what replaced the leads it was generating, and is the new channel actually cheaper per acquisition?