Published October 2026 · Data collected October 3-4, 2026 · Methodology and limits
We looked at 81 content, SEO and digital-PR agencies and scored the 402 most recent posts on their own blogs (840,324 words in all). The aim was a plain picture of how often agency writing shows the phrasing readers now tie to AI tools. No agency is named here, and the scores describe style, not who wrote a post.
| Score band | Posts | Share | Agencies (by average) |
|---|---|---|---|
| Reads human (0-19) | 314 | 78% | 64 |
| A little sloppy (20-44) | 80 | 20% | 17 |
| Sloppy (45-69) | 8 | 2% | 0 |
| Pure slop (70-100) | 0 | 0% | 0 |
Agency average = mean score of that agency's 3-5 sampled posts. Median agency average: 9.
| Pattern | Posts with it | Share of posts | Per 10,000 words |
|---|---|---|---|
| Forced triad | 259 | 64% | 19.3 |
| Em dash | 227 | 56% | 23 |
| Overused AI word | 195 | 49% | 7.6 |
| Not X, but Y | 138 | 34% | 2.7 |
| Shallow -ing rider | 62 | 15% | 1.3 |
| Emoji/arrow decoration | 31 | 8% | 1.5 |
| Row of dramatic fragments | 30 | 7% | 0.6 |
| Borrowed authority | 30 | 7% | 0.4 |
| Avoiding is/has | 29 | 7% | 0.4 |
| Sales language | 21 | 5% | 0.3 |
| Inflated significance | 14 | 3% | 0.2 |
| Hedging opener | 13 | 3% | 0.2 |
| Arguing with no one | 10 | 2% | 0.1 |
| Fake-profound ending | 9 | 2% | 0.1 |
Em dashes and lists of three are weak signals: SlopScore only counts them toward a score when they are dense or appear alongside other patterns. Many careful human writers use both.
crucial · 94 leverage · 43 leveraging · 34 seamless · 34 streamline · 27 showcase · 27 holistic · 24 seamlessly · 20 enhance your · 18 enhance the · 16 showcasing · 14 streamlined · 10
Counts are across all 402 posts. These words are fine on their own; they stand out when they pile up. See words ChatGPT overuses.
| Length | Posts | Median score | Scored 20+ |
|---|---|---|---|
| under 800 words | 67 | 9 | 27% |
| 800 to 1,500 words | 101 | 8 | 23% |
| over 1,500 words | 234 | 6.5 | 20% |
| Year (from page metadata) | Posts | Median score |
|---|---|---|
| 2024 or earlier | 110 | 6 |
| 2025 | 54 | 18.5 |
| 2026 | 199 | 5 |
We took the posts each sitemap listed as most recently updated. Some of those were first published years ago, which is why older dates appear. Dates come from page metadata and aren't always accurate, so read this table loosely.
The patterns in this study are easy to fix once someone points at them. In the 88 posts that scored 20 or more, the two most frequent patterns in each post accounted for a median of 87% of its flags. Usually that means one or two habits repeated, such as the same stock words or a string of em dashes, which an editor can clear quickly. The harder part is doing it on every piece, for every client, and being able to show that a person checked the work.
Two things help. First, a written policy for when AI is allowed and who reviews it. We published a free AI content policy template for agencies you can copy. Second, a quick check before delivery. You can paste any draft into the free SlopScore checker, or score a page by URL.
We started from a list of 100 agencies that sell content, SEO or digital PR. For each one we read robots.txt, found recent posts through the site's sitemap or blog index, and scored 3-5 posts of at least 250 words. 0 sites blocked automated requests or asked crawlers to stay away, and 19 had too few readable posts (or none we could reach), so 81 made it into the index. Full details, the scoring formula and the limits are on the methodology page.
We built a private audit for each agency in the sample, with per-post scores and the flagged lines. If you run an agency and want yours, or want us to run it on client work, email slopscore@agentmail.to. The Agency plan is $99/mo, with a founding price of $49/mo for the first 25 agencies.
Cite as: SlopScore, "Agency AI Writing Index 2026", https://slopscore-nine.vercel.app/research/agency-ai-writing-index-2026/. Aggregate numbers may be reused with a link.