The Originality Premium

The traditional utility of blogging is being flattened by AI. The original intent of blogs — to inform, entertain, and capture search traffic — doesn't produce the same returns anymore. The volume play is finished. What replaces it is narrower and harder. Content earns its place when it is extractable, attributable, and genuinely original. Your voice, your numbers, and your verifiable sourcing are the mechanism that helps AI decide whether to reference you, or not.

Two Directions: Volume and Value

Volume: the content flood. The flood has scale but not substance. Some agencies offer several articles a month, or several a day. That's fine when a human writes them. Most are generated, offer no new perspective, and share the same homogenous language patterns.

Pew Research estimates that roughly 35% of content online shows signs of AI authorship. Graphite went further, analyzing more than 65,000 randomly selected articles of at least 100 words and finding that over half of newly published articles on the web are now machine-generated.

Value: the originality premium. The originality premium is the rising value of unique human perspective, primary data, and lived experience. AI can produce articles at any volume, but the value of that information is short-lived. Answer engines reward content that brings something to the table. The premium is on having an original idea and something specific to say.

What Actually Gets Measured

Data integrity. Real numbers carry weight. Researching this piece, I went deep into the source material to give you more facts than feelings. Firsthand measurement is worth more than topicality. Demonstrable work is the new gold standard.

The "why" factor. Describing what happened is commodity work. Articulating why it happened, in your own voice, is where the value lives. Word choice matters as much as insight.

This isn't just a marketer's hunch. Francesco Agniellini, who teaches Digital and Data Studies at Binghamton, looked at the same Graphite numbers and noticed what kind of writing the machines had taken over: news updates, how-to guides, lifestyle posts, reviews, product explainers. Work whose purpose is to inform or persuade, not to express anything. He also points to research showing these systems pull writers toward similar patterns of wording, flattening the differences that mark an individual voice. His conclusion is the same as mine, arrived at from the other direction: as the flood rises, writing with actual voice and stylistic intention becomes more meaningful, not less.

Distinctive phrasing survives. It becomes the thing that gets quoted, and when it gets misquoted, it makes the origin easier to trace back.

The Supporting Research

The academic work is dry but deep. GEO: Generative Engine Optimization (Aggarwal et al., KDD 2024), out of Princeton and IIT Delhi, tested nine content strategies across roughly 10,000 queries. The interesting part takes some digging: adding source citations produced a 115.1% visibility increase for a site sitting in the fifth position, while the top-ranked site lost 30.3% of its share in the same test. Keyword stuffing barely moved the needle. Statistics and quotations did. Smaller publishers can win here precisely because original, measurable content is what the model is grading.

Search Engine Land ran a companion experiment with SE Ranking worth reading alongside it. They bought 20 new domains with no backlinks, no authority, and no search history, then published 100 unedited AI articles on each — 2,000 articles in total — and walked away. Indexing came fast: about 71% of pages indexed within 36 days, generating 122,102 impressions and 244 clicks in month one. By month three, impressions had climbed past 526,000. Then, in a single week, it collapsed. The share of pages ranking in the top 100 fell from 28% to 3%, and 13 months later almost nothing had recovered.

The Research That Complicates It

Pratyush Kumar's work on brand visibility found that cumulative advantage still holds. What I'd question is whether the ubiquitous brands scored well for having the most content, or simply for being global brands with enormous natural reach before any of this started.

Domain authority also still counts, and there's a number on it. Semrush and Kevin Indig analyzed 1,000 domains across five AI surfaces in October 2025 and found a domain's Authority Score correlated with how often it appeared in AI answers at a Pearson coefficient of 0.65. The important wrinkle is that the effect is threshold-based. Below a minimum authority level, adding links returns close to nothing. Above it, authority predicts citation frequency well. That's a gate, not a ladder.

The gate can also be gamed. One study invented a fake brand, promoted it on aged domains, and it worked. Search Engine Land tested whether classic techniques could prop up a new brand inside AI search, and they could. Tally another point for the skeptics.

And llms.txt adoption sits around 10%, with Google saying it ignores the file entirely. That matters because the promise of llms.txt was letting AI reach a site's best content without fighting through the design.

How Both Are True

Two different things are being measured: retrieval and selection.

Retrieval is the gate. Before anything can be quoted, it has to be indexed, crawled, validated, structured, and backed by enough domain authority to clear the threshold. That layer still rewards age, links, and scale — which is exactly why the fake brand experiment worked, and exactly what the 0.65 is measuring.

Selection is what happens inside that gate. Once several sources are in the retrieval set, the model decides which one to lean on. That decision runs on substance: statistics, sourcing, specificity, clear prose. This is where the GEO findings live, and it's why a fifth-place site can outdraw the leader.

The fake brand cleared the gate. It didn't survive the room. Manufactured authority looks real at launch, but answer engines track corroboration over time — a brand that's genuinely arriving starts appearing in sources its owner doesn't control. When that never happens, the signal decays. Volume gets you considered. It doesn't get you chosen.

Strategic Integration

Keyword stuffing is gone, and article stuffing won't solve this either. It's a short-term fix to a long-term opportunity: do the work, provide value, and be clear.

Working through questions like this one is why we built Full Service Review. FSR makes blog content and broader business data intelligently available to answer engines and humans alike — structured so both can find what matters, specific enough to be quotable, and instrumented so you can see which agents are actually reading it.

Open Questions

What happens to the selection advantage once everyone adopts it? GEO tactics work partly because most content still has nothing concrete in it. If statistics and sourcing become table stakes, the premium moves somewhere else. Where?

How much of the retrieval layer can a small publisher actually influence, given that 0.65 correlation with authority and a demonstrated vulnerability to multi-domain corroboration?

And what's the right way to measure citation independent of clicks? Answer-engine citations are brand impressions that often never produce a visit, and the industry has no agreed unit of account for them.

Sources and Recommended Reading

Aggarwal et al., GEO: Generative Engine Optimization, KDD 2024. https://arxiv.org/abs/2311.09735

Francesco Agniellini, "More than half of new articles on the internet are being written by AI," The Conversation, January 2026 (republished by Binghamton University). https://www.binghamton.edu/news/story/6008/articles-written-by-ai-study

Graphite, "More articles are now created by AI than humans." https://graphite.io/five-percent/more-articles-are-now-created-by-ai-than-humans

Pew Research Center, "How much of the internet is written with AI?" https://www.pewresearch.org/data-labs/2026/08/20/how-much-of-the-internet-is-written-with-ai/

Joe Tidy, on AI slop and the coming backlash, BBC News. https://www.bbc.com/news/articles/c9wx2dz2v44o

SE Ranking / Search Engine Land, 16-month AI content experiment. https://searchengineland.com/ai-generated-content-google-search-experiment-472234

Yotam Gil, "AI made originality the only thing you should pay for," Fast Company, September 8, 2026. https://www.fastcompany.com/91602510/ai-made-originality-the-only-thing-you-should-pay-for

Pratyush Kumar, brand visibility across generative engines. https://arxiv.org/abs/2606.20065

"Fake brands really can win in AI search," Search Engine Land. https://searchengineland.com/fake-brand-ai-search-experiment-475947