I Gave 6 AI Blog Writers the Same Brief: Good Prose Wasn’t Enough
You publish a well-written post. It reads clean, it is accurate, it took effort. And AI assistants still never cite it. That was the brief I wrote for the test: “Why Your Best Blog Posts Got Zero AI Citations.”
I gave six AI blog writers that same brief and had each produce one full article. The six were the Botric Content Agent, PromptWatch, Writesonic, Machined.ai, Byword and AI Clicks.
A disclosure before the results: I work on content at Botric, and the Botric Content Agent is one of the six. It ranked first in my scoring, which is exactly why you should read the weaknesses I found in it too.
How I scored
Four equal categories, 25% each: content quality, research and accuracy, rankability signals, and commercial value. I scored writing out of 10, counted structure against a list of 14 signals, and gave directional estimates for Google ranking and LLM citation likelihood. Two honest limits: that is one article per tool from one brief, and the scores are not live rank tracking or live AI-platform testing. The LLM citation estimate is inferred from structure, not measured.
The 14 structural signals
I track these because AI systems split a page into passages, score each for relevance and how easily it can be extracted, and cite the passage that answers the question as a self-contained unit:
A named author with credentials. An answer-first opener. Question-based H2s. Source-traced statistics with publication and date. A consolidated sources section. An FAQ section with six or more questions. FAQPage JSON-LD. A comparison table. Article JSON-LD. BreadcrumbList schema. HowTo schema. A key takeaways section. Inline definition boxes. A disclosure statement.
What each tool produced
Botric Content Agent Structure: 14 of 14 signals, including Article, FAQPage, BreadcrumbList and HowTo JSON-LD Writing: 8.0 out of 10 Sources: 10 or more named What I found against it: the prose lacks personality, and its comparison articles give Botric softer limitations than competitors.
Get Botric’s stories in your inbox
Join Medium for free to get updates from this writer.
PromptWatch Structure: 0 of 14 signals Writing: 9.0 out of 10, the best prose of the six, with strong argument construction What I found: it needs a full structural retrofit before it is citation-ready.
Writesonic Structure: 3 of 14 signals Writing: 6.5 out of 10 What I found: 34 numbered references, including Nature, arXiv and Los Alamos National Laboratory, and a good angle on “ghost citations”. But no structure, and the title targeted a different keyword than the brief.
Machined.ai Structure: 0 of 14 signals Writing: 7.5 out of 10, warm and natural prose What I found: no sources, statistics or data.
Byword Structure: 0 of 14 signals Writing: 7.0 out of 10 What I found: a detailed explanation of retrieval-augmented generation, but unsourced statistics.
AI Clicks Structure: 0 of 14 signals Writing: 5.0 out of 10, the weakest in the test, in product-marketing language What I found: its comparison table used vague entries, and it cited its own blog as a source, which I flagged as a potential conflict of interest.
What the test actually showed
Writing quality and AI citation readiness are different skills. The best prose in the test belonged to a tool with no structural signals.
Sourcing without structure does not get you there either. Writesonic had the deepest reference list and still shipped only three of the 14 signals.
And the advantage was structural, not linguistic. Structure is what makes sources citable, not the other way around.
What I would do with this
If I were checking one of my own posts today, I would not start by rewriting the prose. I would run it against the 14 signals: is there a named author, an answer-first opener, question-style headings, source-traced statistics, an FAQ with schema, and a disclosure? Then I would fix the gaps.
For where to look first, the guide on what is actually blocking citations is a good companion, and a free GEO audit shows how a site reads to AI engines. The full comparison has every score and the FAQs.
