Your buyers stopped Googling. Now they ask AI who to buy from — and AI answers with names, not links. HarperFlow puts you on that shortlist. Not a writer — an editorial operation: it finds what your buyers ask AI, researches each article with 8–12 dated sources, fact-checks every claim, and publishes into Webflow, WordPress or Shopify with the FAQs, tables and JSON-LD engines quote. 5–30 articles a month in your voice, with veto before it goes live. 7-day trial: 3 articles; card required.
Forum Threads
We ran 100+ AI-citation audits before launching. Here's what actually gets cited.
Before writing a line of product marketing, we audited 100+ real business sites: same buyer questions, asked live to ChatGPT, Perplexity and Gemini, then we read every page the engines cited. What the cited pages have in common:
77% are list/comparison formats ("best X", "X vs Y") the format wins before the brand does
only ~31% carry original data (tests, benchmarks, surveys) which is wild, because engines love citing numbers. Most sites simply don't publish any numbers or original data.
named authors and real depth (~3,000+ words on cited pages) show up constantly and on Perplexity, Reddit is the most-cited source on ~87% of questions
The uncomfortable summary: most companies aren't invisible to AI because they're bad they're invisible because their content isn't in the formats AI quotes. That's the gap we built HarperFlow for: a blog-on-autopilot engine, built to be cited every article fact-grounded with 8 12 cited sources and ~13 20 minutes of live research by the pipeline, published straight into your CMS.
I'm Hesham, the solo builder behind HarperFlow.
I'm a Cambridge-trained doctor who somehow ended up building content machines. HarperFlow is an AI blogging tool that publishes directly to your Wordpress, Webflow, or Shopify site. It started with an n8n workflow I'd built for a client that I felt could help businesses better leverage their company blogs for GEO/SEO. I spent months working on perfecting HarperFlow to ensure that it doesn't just end up being one of those low quality spammy article generators.
Instead, we create articles that are engineered to be citable by LLMs.
To make the articles highly citable by LLMs, I made sure each article we publish on a customer's site had JSON-LDs, author schema, H2s/H3s, tags, links to the original sources for every fact, TL;DR Takeaways, internal linking, FAQ blocks for each article (and so on... it's a long list). Naturally, we started using our tool on our own website and we've been able to get some decent traction; primarily through our blog... especially for a domain that's 100 days old or so.
Even better... this has helped improve our DA better than any backlinks could, which (to be honest) wasn't something that I was necessarily expecting. So far, we've managed 13,000+ citations on our own site using our own tool.
You can use this coupon code HF-JPG6G-BKX5U to get 50% off on your first 3 months.
Or, if you don't want to commit just yet... you can use HF-XHVCP-MXF29 for 5 free articles.
Please let me know if you have any questions (or would like to be onboarded 1:1).
— Hesham
@heshamus wow seeing 13,000 plus citations on your own site is probably the strongest proof here, especially for such a young domain. Plenty of tools talk about getting mentioned in LLM answers, but actually using the product on your own site and showing the results makes the story much more convincing.
This was interesting since I have been trying to get any traction with our company blog
skeptical of the causal chain here specifically: JSON-LD, author schema, and FAQ blocks are things Google's crawler and its structured-data pipeline explicitly parse, but I don't think that's how LLM citation actually works. answer engines like Perplexity cite what their live web search retrieves and ranks, and that ranking still leans heavily on classic SEO signals (backlinks, domain authority, content relevance), not schema markup the model itself reads. is the "100 days old, decent traction" result separable from just... publishing good, well-structured content that would've ranked in Google too, or is there evidence the schema specifically is doing work in LLM citations
@galdayan We don't currently have evidence that the schema is doing work in LLM citations, but what I will say is that we do capture DA and all that other stuff backstage... and find keyword opportunities that are suitable for that site's DA and its traffic and all that. So, we won't be picking a keyword opportunity that has a KD of 80, for example, when the site is still very young or has limited DA. On a separate note, our research suggests that FAQ blocks and tables do uplift AI citations (p=0.0004) and app does publish those as art of the articles.