Skip to content

The Lab

“In today’s fast-paced digital landscape,” Do AI models still open this way?

The newest models have almost stopped, and they have picked up new habits instead. One line keeps coming back: “Running a small business means wearing a lot of hats.”

The exception is ChatGPT itself. It opened more than half of its test intros with “In today’s”, while GPT-6 through OpenAI’s API almost never did.

By Alex Rostovtsev and Kristina Tyumeneva, 6 November 2026. This is the data behind Fireplace Stories Vol.1, published on LinkedIn.

Every intro, one square each

Each row is a model or app. Each column is one of ten business topics, asked three times. Hover over a square or tap it to read how that intro began, and open the full answer from there.

91 of 450 intros opened with “In today’s”.

ChatGPT, Staying competitive, first run

In today’s fast-paced digital landscape, staying competitive online is more important than ever.

Model
Digital marketingContent marketingStaying competitiveSEOEmail marketingCustomer serviceRemote teamsCybersecuritySocial mediaWebsite speed
Lit
GPT-4o mini
27
GPT-4.1
26
GPT-5
0
Claude Opus 4.5
11
Gemini 2.5 Flash
8
GPT-6 Sol
0
GPT-6 Luna
2
GPT-6 Astra
0
GPT-6.1 Sol
0
Claude Opus 5.5
0
Claude Sonnet 5.5
0
Gemini 4 Argon
0
Grok 4.7
0
ChatGPT
17
Gemini
0

What the test shows

Older models still reach for it

GPT-4o mini and GPT-4.1 opened this way on all ten topics, 53 times out of 60 between them. Claude Opus 4.5 and Gemini 2.5 Flash did it less often, and GPT-5 had already stopped in 2025.

The full line is rarer than its reputation. “In today’s fast-paced digital landscape” appeared word for word in 22 of the 450 opening sentences, and 12 of those came from GPT-4o mini.

Current models have nearly dropped it

The eight current models we called through their APIs wrote 240 intros. Two of them opened with “In today’s”, both from GPT-6 Luna, and none used the full line.

ChatGPT hasn’t dropped it

ChatGPT opened with “In today’s” 17 times out of 30, and DataForSEO reports the model behind it as GPT-6. Through the API, GPT-6 did that twice in 90 intros. A gap like that could have been a bad day, so we ran ChatGPT again the next day, from the US and from Australia. The count barely moved.

We can’t see inside ChatGPT, so we can’t say why. The raw answers do show that the product adds something the API doesn’t: 28 of those 30 arrived wrapped in a writing block. That layer may be where the old opener comes from, though this test can’t prove it.

The new stock openers

A stock opener is a sentence that could start any article on the subject. The old one has faded, and the newest models have found others. We noticed these while reading the answers, so they are observations and the test wasn’t designed to measure them. Selecting one lights its matches on the wall.

  • Running a small business means wearing a lot of hats

    GPT-6 Sol, Astra and 6.1 Sol, in 13 of their 27 small-business intros

    “Running a small business means wearing a lot of hats—and finding time to market it can feel like one task too many.”

    Light it up on the wall

  • takes more than

    All four GPT-6 models, 12 of 120 intros, and no older model

    “Remote work gives teams more flexibility, but staying productive takes more than a good Wi-Fi connection.”

    Light it up on the wall

  • Remember when

    Gemini 4 Argon, 8 of 30

    “Remember when simply having a website was enough to set your business apart?”

    Light it up on the wall

  • Picture this:

    Claude Opus 5.5, 5 of 30

    “Picture this: a potential customer in your town pulls out their phone and searches for exactly what you sell.”

    Light it up on the wall

  • Ten years ago,

    Claude Sonnet 5.5, 5 of 30

    “Ten years ago, a small business could build a loyal customer base with a good location, a friendly smile, and a few well-placed flyers.”

    Light it up on the wall

  • If you run a small business,

    GPT-5, 7 of 30

    “If you run a small business, “digital marketing” can sound like a buzzword reserved for big brands with bigger budgets.”

    Light it up on the wall

Different models also arrive at the same sentence. “A slow website can lose a visitor before they’ve read a single word.” appeared word for word 5 times, from GPT-6 Astra and GPT-6.1 Sol.

The em dash, the tell many readers look for, turned out to be a family trait. GPT-6.1 Sol used one in 19 of its 30 opening sentences, and neither Claude 5.5 model used one at all.

Where the phrase came from

The line is older than any of these models. Google Books finds “fast-paced digital” from the mid-1990s, and its use climbs most steeply after 2017, well before ChatGPT launched.

In the 2000s, content farms paid writers by the article. With little to say and a word count to reach, a stock opener was the fastest way in. In 2010, Yahoo spent around $100 million on one of them, Associated Content. Google pushed those sites down the results with the Panda update in 2011, but the habit carried on in corporate copy. Ten years later, a Business Wire release was still using “In today’s fast-paced digital economy”.

Most of what GPT-3 trained on came from a filtered scrape of the open web, according to OpenAI’s paper. That web had spent years opening articles this way.

19962022
How often the words turned up in books each year. The data stops in 2022.

How the test worked

On 5 November 2026 we gave 15 models and apps the same ten requests for a blog intro and ran each one three times. That came to 450 answers for USD 1.54. The prompts carried no system message or style guide. The API models ran with web search off, and each provider got one request every two seconds.

The prompts

  1. Write an intro for a blog post about digital marketing for small businesses.
  2. Write the opening paragraph of a blog post about why companies should invest in content marketing.
  3. Write a blog intro about how businesses can stay competitive online.
  4. Write an intro for a blog post about SEO for small businesses.
  5. Write the opening paragraph of a blog post about email marketing.
  6. Write a blog intro about customer service trends.
  7. Write an intro for a blog post about keeping remote teams productive.
  8. Write the opening paragraph of a blog post about cybersecurity for small businesses.
  9. Write a blog intro about using social media to grow a business.
  10. Write an intro for a blog post about why website speed matters.

How an answer was scored

Only one sentence counts: the first sentence of the intro itself. The script skips titles, preambles such as “Here’s an intro” and option labels. When a model offers several versions, only the first is scored, so every model is judged on one intro. The rules were written before the full run.

Opened with “In today’s”
The sentence starts with those words.
The exact phrase
The sentence contains “In today’s fast-paced digital landscape”.
“Fast-paced digital”
Those words, with or without one word between them, as in “fast-paced, ever-evolving digital”.
Era opener
Any of the above, or a first sentence leaning on “digital age”, “digital era”, “digital landscape”, “digital world”, “ever-evolving”, “rapidly changing” or “fast-paced”, or opening “In the modern” or “In an era”.

Results

Out of 30 intros per row. Topics counts how many of the ten topics opened with “In today’s” at least once.
Model In today’s Topics Exact Fast-paced Era
Older models
GPT-4o mini 2024 27 10 12 15 29
GPT-4.1 2025 26 10 4 9 26
GPT-5 2025 0 0 0 0 1
Claude Opus 4.5 2025 11 4 2 2 15
Gemini 2.5 Flash 2025 8 4 1 2 12
Current models
GPT-6 Sol 0 0 0 0 0
GPT-6 Luna 2 2 0 0 3
GPT-6 Astra 0 0 0 0 0
GPT-6.1 Sol 0 0 0 0 0
Claude Opus 5.5 0 0 0 0 0
Claude Sonnet 5.5 0 0 0 0 0
Gemini 4 Argon 0 0 0 0 5
Grok 4.7 0 0 0 0 0
Inside the apps
ChatGPT 17 6 3 6 17
Gemini 0 0 0 0 3

How each model was reached

Model Reached through Reported as
GPT-4o mini DataForSEO LLM Responses API gpt-4o-mini-2024-07-18
GPT-4.1 DataForSEO LLM Responses API gpt-4.1-2025-04-14
GPT-5 DataForSEO LLM Responses API gpt-5-2025-08-07
Claude Opus 4.5 Anthropic API claude-opus-4-5-20251101
Gemini 2.5 Flash DataForSEO LLM Responses API gemini-2.5-flash
GPT-6 Sol DataForSEO LLM Responses API gpt-6-sol
GPT-6 Luna DataForSEO LLM Responses API gpt-6-luna
GPT-6 Astra DataForSEO LLM Responses API gpt-6-astra
GPT-6.1 Sol DataForSEO LLM Responses API gpt-6.1-sol
Claude Opus 5.5 Anthropic API claude-opus-5-5
Claude Sonnet 5.5 Anthropic API claude-sonnet-5-5
Gemini 4 Argon DataForSEO LLM Responses API gemini-4-argon
Grok 4.7 xAI API grok-4.7
ChatGPT DataForSEO LLM Scraper, US location gpt-6
Gemini DataForSEO LLM Scraper, US location 3.5 Flash-Lite

The repeat run

On 6 November we repeated the two app rows with the same prompts. ChatGPT from the US opened with “In today’s” 18 times out of 30, and ChatGPT from Australia 17. Gemini went from 0 on the first day to 3, all three on the staying-competitive topic. These 90 answers sit in their own file and aren’t counted in the 450.

Limits

  • The test covers two days, ten business topics and default settings. It shows how these models open a marketing-style blog post when nobody tells them how to write, and says little about how they write in general.
  • Three runs of one prompt aren’t independent. ChatGPT’s 17 covers 6 of the 10 topics because it often repeats itself, which is why the results table counts topics too.
  • The app rows came through DataForSEO’s scraper with a US location, not from a person typing. DataForSEO describes its ChatGPT route as search mode, so a logged-in chat may behave differently. None of ChatGPT’s answers carried links, while 8 of Gemini’s 30 did, so Gemini was drawing on the web for at least those.
  • Model names are as each provider or DataForSEO reported them. We can’t verify what runs behind an app.
  • GPT-6 Sol, Luna and Astra were added after the first scoring, to test the ChatGPT result. The scoring rules didn’t change.
  • “In today’s” also shows up further down 32 answers that didn’t open with it, often in a second or third version of the intro. Those don’t count, by the rule above.
  • The new stock openers came from reading the answers after the run, and they need a test of their own.
  • One Gemini answer had a citation link attached to its first sentence, and the script first read that as one long sentence. We corrected the script on 6 November, and no score changed.
  • We built the test scripts and this page with Claude Code, and Anthropic’s models are among those tested. The scoring is a fixed script, so no model judged any answer.

The data

Every answer is here with its prompt, scores and cost. You can reuse it under CC BY 4.0, with a credit to this page.

To cite it: Alex Rostovtsev and Kristina Tyumeneva, “Do AI models still write ‘In today’s fast-paced digital landscape’?”, BLURSOR, 6 November 2026.