@font-face {
font-family: ‘Guardian Headline Full’;
src: url(‘https://assets.guim.co.uk/static/frontend/fonts/guardian-headline/noalts-not-hinted/GHGuardianHeadline-Light.woff2’) format(‘woff2’),
url(‘https://assets.guim.co.uk/static/frontend/fonts/guardian-headline/noalts-not-hinted/GHGuardianHeadline-Light.woff’) format(‘woff’),
url(‘https://assets.guim.co.uk/static/frontend/fonts/guardian-headline/noalts-not-hinted/GHGuardianHeadline-Light.ttf’) format(‘truetype’);
font-weight: 300;
font-style: normal;
}
@font-face {
font-family: ‘Guardian Headline Full’;
src: url(‘https://assets.guim.co.uk/static/frontend/fonts/guardian-headline/noalts-not-hinted/GHGuardianHeadline-LightItalic.woff2’) format(‘woff2’),
url(‘https://assets.guim.co.uk/static/frontend/fonts/guardian-headline/noalts-not-hinted/GHGuardianHeadline-LightItalic.woff’) format(‘woff’),
url(‘https://assets.guim.co.uk/static/frontend/fonts/guardian-headline/noalts-not-hinted/GHGuardianHeadline-LightItalic.ttf’) format(‘truetype’);
font-weight: 300;
font-style: italic;
}
@font-face {
font-family: ‘Guardian Headline Full’;
src: url(‘https://assets.guim.co.uk/static/frontend/fonts/guardian-headline/noalts-not-hinted/GHGuardianHeadline-Regular.woff2’) format(‘woff2’),
url(‘https://assets.guim.co.uk/static/frontend/fonts/guardian-headline/noalts-not-hinted/GHGuardianHeadline-Regular.woff’) format(‘woff’),
url(‘https://assets.guim.co.uk/static/frontend/fonts/guardian-headline/noalts-not-hinted/GHGuardianHeadline-Regular.ttf’) format(‘truetype’);
font-weight: 400;
font-style: normal;
}
@font-face {
font-family: ‘Guardian Headline Full’;
src: url(‘https://assets.guim.co.uk/static/frontend/fonts/guardian-headline/noalts-not-hinted/GHGuardianHeadline-RegularItalic.woff2’) format(‘woff2’),
url(‘https://assets.guim.co.uk/static/frontend/fonts/guardian-headline/noalts-not-hinted/GHGuardianHeadline-RegularItalic.woff’) format(‘woff’),
url(‘https://assets.guim.co.uk/static/frontend/fonts/guardian-headline/noalts-not-hinted/GHGuardianHeadline-RegularItalic.ttf’) format(‘truetype’);
font-weight: 400;
font-style: italic;
}
@font-face {
font-family: ‘Guardian Headline Full’;
src: url(‘https://assets.guim.co.uk/static/frontend/fonts/guardian-headline/noalts-not-hinted/GHGuardianHeadline-Medium.woff2’) format(‘woff2’),
url(‘https://assets.guim.co.uk/static/frontend/fonts/guardian-headline/noalts-not-hinted/GHGuardianHeadline-Medium.woff’) format(‘woff’),
url(‘https://assets.guim.co.uk/static/frontend/fonts/guardian-headline/noalts-not-hinted/GHGuardianHeadline-Medium.ttf’) format(‘truetype’);
font-weight: 500;
font-style: normal;
}
@font-face {
font-family: ‘Guardian Headline Full’;
src: url(‘https://assets.guim.co.uk/static/frontend/fonts/guardian-headline/noalts-not-hinted/GHGuardianHeadline-MediumItalic.woff2’) format(‘woff2’),
url(‘https://assets.guim.co.uk/static/frontend/fonts/guardian-headline/noalts-not-hinted/GHGuardianHeadline-MediumItalic.woff’) format(‘woff’),
url(‘https://assets.guim.co.uk/static/frontend/fonts/guardian-headline/noalts-not-hinted/GHGuardianHeadline-MediumItalic.ttf’) format(‘truetype’);
font-weight: 500;
font-style: italic;
}
@font-face {
font-family: ‘Guardian Headline Full’;
src: url(‘https://assets.guim.co.uk/static/frontend/fonts/guardian-headline/noalts-not-hinted/GHGuardianHeadline-Semibold.woff2’) format(‘woff2’),
url(‘https://assets.guim.co.uk/static/frontend/fonts/guardian-headline/noalts-not-hinted/GHGuardianHeadline-Semibold.woff’) format(‘woff’),
url(‘https://assets.guim.co.uk/static/frontend/fonts/guardian-headline/noalts-not-hinted/GHGuardianHeadline-Semibold.ttf’) format(‘truetype’);
font-weight: 600;
font-style: normal;
}
@font-face {
font-family: ‘Guardian Headline Full’;
src: url(‘https://assets.guim.co.uk/static/frontend/fonts/guardian-headline/noalts-not-hinted/GHGuardianHeadline-SemiboldItalic.woff2’) format(‘woff2’),
url(‘https://assets.guim.co.uk/static/frontend/fonts/guardian-headline/noalts-not-hinted/GHGuardianHeadline-SemiboldItalic.woff’) format(‘woff’),
url(‘https://assets.guim.co.uk/static/frontend/fonts/guardian-headline/noalts-not-hinted/GHGuardianHeadline-SemiboldItalic.ttf’) format(‘truetype’);
font-weight: 600;
font-style: italic;
}Here’s the rewritten version in fluent, natural English:
“`css
@font-face {
font-family: ‘Guardian Headline Full’;
src: url(‘https://assets.guim.co.uk/static/frontend/fonts/guardian-headline/noalts-not-hinted/GHGuardianHeadline-Bold.woff2’) format(‘woff2’),
url(‘https://assets.guim.co.uk/static/frontend/fonts/guardian-headline/noalts-not-hinted/GHGuardianHeadline-Bold.woff’) format(‘woff’),
url(‘https://assets.guim.co.uk/static/frontend/fonts/guardian-headline/noalts-not-hinted/GHGuardianHeadline-Bold.ttf’) format(‘truetype’);
font-weight: 700;
font-style: normal;
}
@font-face {
font-family: ‘Guardian Headline Full’;
src: url(‘https://assets.guim.co.uk/static/frontend/fonts/guardian-headline/noalts-not-hinted/GHGuardianHeadline-BoldItalic.woff2’) format(‘woff2’),
url(‘https://assets.guim.co.uk/static/frontend/fonts/guardian-headline/noalts-not-hinted/GHGuardianHeadline-BoldItalic.woff’) format(‘woff’),
url(‘https://assets.guim.co.uk/static/frontend/fonts/guardian-headline/noalts-not-hinted/GHGuardianHeadline-BoldItalic.ttf’) format(‘truetype’);
font-weight: 700;
font-style: italic;
}
@font-face {
font-family: ‘Guardian Headline Full’;
src: url(‘https://assets.guim.co.uk/static/frontend/fonts/guardian-headline/noalts-not-hinted/GHGuardianHeadline-Black.woff2’) format(‘woff2’),
url(‘https://assets.guim.co.uk/static/frontend/fonts/guardian-headline/noalts-not-hinted/GHGuardianHeadline-Black.woff’) format(‘woff’),
url(‘https://assets.guim.co.uk/static/frontend/fonts/guardian-headline/noalts-not-hinted/GHGuardianHeadline-Black.ttf’) format(‘truetype’);
font-weight: 900;
font-style: normal;
}
@font-face {
font-family: ‘Guardian Headline Full’;
src: url(‘https://assets.guim.co.uk/static/frontend/fonts/guardian-headline/noalts-not-hinted/GHGuardianHeadline-BlackItalic.woff2’) format(‘woff2’),
url(‘https://assets.guim.co.uk/static/frontend/fonts/guardian-headline/noalts-not-hinted/GHGuardianHeadline-BlackItalic.woff’) format(‘woff’),
url(‘https://assets.guim.co.uk/static/frontend/fonts/guardian-headline/noalts-not-hinted/GHGuardianHeadline-BlackItalic.ttf’) format(‘truetype’);
font-weight: 900;
font-style: italic;
}
@font-face {
font-family: ‘Guardian Titlepiece’;
src: url(‘https://assets.guim.co.uk/static/frontend/fonts/guardian-titlepiece/noalts-not-hinted/GTGuardianTitlepiece-Bold.woff2’) format(‘woff2’),
url(‘https://assets.guim.co.uk/static/frontend/fonts/guardian-titlepiece/noalts-not-hinted/GTGuardianTitlepiece-Bold.woff’) format(‘woff’),
url(‘https://assets.guim.co.uk/static/frontend/fonts/guardian-titlepiece/noalts-not-hinted/GTGuardianTitlepiece-Bold.ttf’) format(‘truetype’);
font-weight: 700;
font-style: normal;
}
@media (min-width: 71.25em) {
.content__main-column–interactive {
margin-left: 160px;
}
}
@media (min-width: 81.25em) {
.content__main-column–interactive {
margin-left: 240px;
}
}
.content__main-column–interactive .element-atom {
max-width: 620px;
}
@media (max-width: 46.24em) {
.content__main-column–interactive .element-atom {
max-width: 100%;
}
}
.content__main-column–interactive .element-showcase {
margin-left: 0;
}
@media (min-width: 46.25em) {
.content__main-column–interactive .element-showcase {
max-width: 620px;
}
}
@media (min-width: 71.25em) {
.content__main-column–interactive .element-showcase {
max-width: 860px;
}
}
.content__main-column–interactive .element-immersive {
max-width: 1100px;
}
@media (max-width: 46.24em) {
.content__main-column–interactive .element-immersive {
width: calc(100vw – var(–scrollbar-width, 0px));
position: relative;
left: 50%;
right: 50%;
margin-left: calc(-50vw + var(–half-scrollbar-width, 0px)) !important;
margin-right: calc(-50vw + var(–half-scrollbar-width, 0px)) !important;
}
}
@media (min-width: 46.25em) {
.content__main-column–interactive .element-immersive {
transform: translate(-20px);
width: calc(100% + 60px);
}
}
@media (max-width: 71.24em) {
.content__main-column–interactive .element-immersive {
margin-left: 0;
margin-right: 0;
}
}
@media (min-width: 71.25em) {
.content__main-column–interactive .element-immersive {
transform: translate(0);
width: auto;
}
}
@media (min-width: 81.25em) {
.content__main-column–interactive .element-immersive {
max-width: 1260px;
}
}
.content__main-column–interactive p,
.content__main-column–interactive ul {
max-width: 620px;
}
.content__main-column–interactive:before {
position: absolute;
top: 0;
height: calc(100% + 15px);
min-height: 100px;
content: “”;
}
@media (min-width: 71.25em) {
/ Additional styles can go here /
}
“`Here’s the rewritten CSS in fluent, natural English:
The main content column for interactive pages has a left border that is 1 pixel solid light gray. On screens wider than 81.25em, this border shifts slightly to the left. Inside this column, elements like atoms have no top or bottom margin but have 12 pixels of padding on each side. When a paragraph is followed by an atom, the padding is removed and replaced with 12 pixels of margin on both sides. Inline elements are limited to a maximum width of 620 pixels, and on screens wider than 61.25em, figures with the inline role also have this width limit.
For media sections containing a loop figure, the caption sits above other content with a z-index of 6. The loop button is 32 pixels wide, aligned to the bottom right, with a small right margin and extra bottom space. The caption button has a high z-index to stay on top. On screens wider than 46.25em, cinemagraph figures inside media sections have no maximum height restriction.
In the body section, self-hosted videos are displayed as block elements, up to 620 pixels wide, with 12 pixels of margin on top and bottom. The loop figure and its video inside take full width, auto height, and are centered. If the loop figure has the immersive video class, it stretches to full width with no margin. On screens wider than 71.25em, this immersive video expands to 1140 pixels wide and shifts left by 180 pixels, with its caption indented. On screens wider than 81.25em, it grows to 1300 pixels wide and shifts left by 260 pixels.
The root variables define colors for dateline, header border, caption text, caption background, and feature. The feature color is red, and the new pillar color defaults to the primary pillar or feature. Subheading, pullquote text, pullquote icon, and block quote text all use the secondary pillar color. Block quotes also use the secondary pillar for their fill. In dark mode, these colors switch to dark mode pillar values, unless the light color scheme is explicitly set.
Interactive main column elements and atom elements have no padding. In various content areas, the first atom or horizontal rule is followed by a paragraph with no extra spacing. This applies to article body, interactive content, comment body, feature body, and the body data attribute.Here’s the rewritten text in fluent, natural English:
The first paragraph after an element atom, sign-in gate, or horizontal rule (except the last one) gets 14 pixels of padding at the top. The first letter of that paragraph uses the Guardian Headline font family, is bold, 111 pixels tall, and has a line height of 92 pixels. It floats to the left, is uppercase, has 8 pixels of right margin, and uses the drop cap color.
Paragraphs that come right after a horizontal rule have no top padding. Pull quotes are limited to a maximum width of 620 pixels.
For showcase images, the caption is normally positioned statically and takes up the full width, up to 620 pixels. On wider screens (71.25em and above), the caption becomes absolutely positioned with a max width of 140 pixels. On even wider screens (81.25em and above), the max width increases to 220 pixels.
Immersive elements take up the full viewport width, minus the scrollbar width. On smaller screens (up to 71.24em), they max out at 978 pixels, and their captions have 10 pixels of padding on each side. On screens between 30em and 71.24em, the caption padding increases to 20 pixels.Here’s the rewritten version in fluent, natural English:
The immersive element has a maximum width of 738px. On smaller screens (under 46.24em), the immersive element should have a left margin of -10px and no right margin. On screens between 30em and 46.24em, the left margin increases to -20px, and the caption gets 20px of padding on each side.
For larger screens (71.25em and above), showcase images in the body section should have a left margin of -160px. On very large screens (81.25em and above), that margin increases to -240px.
The furniture wrapper is positioned relatively. On screens 61.25em and wider, it becomes a grid with 10 columns and specific rows for the title, headline, meta, standfirst, and portrait sections. The first child of the headline gets a top border. The meta section has some top padding and no right margin. The standfirst’s first paragraph also gets a top border, and its list items are 20px in size. Links in the standfirst are underlined with a specific offset and color, and they change color on hover.
On screens 71.25em and wider, the grid changes to have 14 columns. The meta section gets a 540px-wide top border. The standfirst’s first paragraph no longer has a top border. Figures in the wrapper have a left margin of -10px, and inline figures have a maximum width of 630px.The `e=standfirst]:before` rule adds a 1px-wide vertical line using the header border color. It’s positioned at the top left, with a height of 100%.
For screens wider than 81.25em, the `.furniture-wrapper` uses a grid layout with specific column and row definitions. The `#meta:before` and `[data-gu-name=meta]:before` elements are 620px wide. The `.standfirst:before`, `#standfirst:before`, and `[data-gu-name=standfirst]:before` elements are shifted slightly to the left.
Inside the article header, the labels in `.content__labels > div` for both `.article-header` and `[data-gu-name=title]` have 2px of padding at the top.
The headline (`#headline h1`, `[data-gu-name=headline] h1`, `.headline h1`) is bold (font-weight 600), has a max width of 620px, and a font size of 32px. On screens wider than 71.25em, the max width shrinks to 540px and the font size increases to 50px.
For screens wider than 46.25em, `.keyline-4` and `[data-gu-name=lines]` have no right margin. On screens wider than 61.25em, they are hidden. Their SVG elements use the header border color as the stroke.
On screens wider than 46.25em, `#meta` and `[data-gu-name=meta]` also have no right margin. The social and comment sections inside them use the header border color for their borders. The `gu-island` elements inside the meta container are hidden.
The standfirst sections (`.standfirst`, `#standfirst`, `[data-gu-name=standfirst]`) are positioned with a left margin of -10px, 10px of left padding, and relative positioning. On screens wider than 46.25em, they get 2px of top padding. Paragraphs inside them have normal weight (400), a font size of 20px, and 14px of bottom padding.
The main media section (`#main-media`, `[data-gu-name=media]`) is positioned relatively, has no top margin, a 2px bottom margin, and is placed in the portrait grid area. Its inner divs are full width with no inline margin. On screens wider than 61.25em, the bottom margin is removed. On screens narrower than 46.24em, it spans the full viewport width (minus scrollbar) and has a left margin of -10px (or -20px on screens wider than 30em).
The figcaption is positioned at the bottom, with padding of 4px on top and sides and 12px at the bottom. It uses the caption background and text colors, has no max width, is full width, no bottom margin, and a minimum height of 46px. The span inside uses the header border color, and its SVG fills that color. The first span is hidden, while the second is shown with a max width of 90%. On screens wider than 30em, the figcaption padding changes to 4px on top and 20px on the sides. When hidden, its opacity is 0.
The caption button (`#caption-button`) is displayed as a block, positioned at the bottom right (10px from bottom, 8px from right), with a z-index of 30. It uses the caption background color, has no border, and is fully rounded.Here is the rewritten text in fluent, natural English:
The CSS code adjusts the layout and styling of article pages, especially for mobile devices like iOS and Android. It sets padding, button sizes, and positioning for captions and interactive content. The design uses custom colors for dark mode and pillar-based themes. On iOS and Android, the first letter of the first paragraph after certain elements is colored based on the secondary pillar. The article header is hidden by setting its height to zero. The furniture wrapper adds padding and styles the content labels with bold, capitalized text in a serif font, using the pillar color. The headline is also adjusted within the furniture wrapper.Here’s the rewritten CSS in fluent, natural English:
For the headline inside the furniture wrapper on Android devices, the font size is 32 pixels, bold, with 12 pixels of padding at the bottom, and the color is set to a dark shade (#121212).
On both iOS and Android, when an image appears inside the furniture wrapper of any article container (feature, standard, or comment), it is positioned relatively. It has a 14-pixel top margin and no left margin, but it starts 10 pixels to the left. Its width takes up the full viewport width, minus any scrollbar width, and its height adjusts automatically.
For the image itself, along with its inner container and any links inside it, the background is transparent. The width is again the full viewport width minus the scrollbar, and the height is set to auto (important).
The standfirst section on both iOS and Android has 4 pixels of padding at the top and 24 pixels at the bottom, with a negative right margin of 10 pixels.
Inside the standfirst, any paragraph text uses the font family: Guardian Headline, Guardian Egyptian Web, Guardian Headline Full, Georgia, or a serif fallback.
Links inside the standfirst, whether in a list item or not, follow the same styling.Here’s the rewritten version in fluent, natural English:
On Android devices, links inside the standfirst section of feature, standard, and comment articles use the pillar color, have no background image, are underlined with a 6px offset, and use a light gray underline color instead of a bottom border.
On both iOS and Android, when you hover over those same links, the underline color changes to the pillar color.
On both platforms, the meta section in these article containers has no margin.
Also on both platforms, the byline text, author names, author links, and any byline spans inside the meta section all use the pillar color.
The same styling applies to the meta__misc section in feature and standard articles on iOS, as well as in comment articles.For iOS and Android, the `.meta__misc` section inside `.furniture-wrapper` within article containers (feature, standard, and comment) has no padding.
On both iOS and Android, the SVG icons in the `.meta__misc` section of these article containers use a stroke color defined by `–new-pillar-colour`.
For iOS and Android, the caption button (`#caption-button`) inside `.element–showcase` within the `.furniture-wrapper` of all article containers is displayed as a flexbox. It has 5px padding, centered content both horizontally and vertically, and is 28px wide and 28px tall, positioned 14px from the right.
On iOS and Android, the article body (`.article__body`) in feature, standard, and comment containers has 12px padding on the left and right, with no top or bottom padding.
For iOS and Android, images inside the article body that are not thumbnails or immersive (i.e., `figure.element-image:not(.element–thumbnail):not(.element–immersive)`) have no margin. Their width is calculated as `100vw – 24px – var(–scrollbar-width, 0px)`, and their height is automatic. The captions for these images have no padding.
For iOS and Android, immersive images (`figure.element-image.element-immersive`) inside the article body have a width of `100vw – var(–scrollbar-width, 0px)`.
On iOS, within the article body, blockquotes with the class `quoted` have a `:before` pseudo-element (the specific styling is not fully provided here).Here’s the rewritten version in fluent, natural English:
For quoted text in articles, the color is set by the pillar color variable. Links within article text are styled with an underline, using the primary pillar color and a 6px offset. The underline color matches the header border. When you hover over these links, the underline changes to the new pillar color.
In dark mode, the furniture wrapper background becomes dark gray (#1a1a1a). The content labels inside it use the new pillar color. The main headline doesn’t have a background and its color is set by the header border. Standfirst text also uses the header border color. Links in the standfirst and author names in the byline follow the same styling rules.Here’s the rewritten version in fluent, natural English:
On both Android and iOS, the byline author links in the meta section of feature, standard, and comment articles use the new pillar color. Similarly, the SVG icons in the meta misc section of these articles are outlined in the same new pillar color.
For showcase images with captions, the caption text color is set to the dateline color on both platforms.
Blockquotes within the article body that are marked as quoted also use the new pillar color on both Android and iOS.
Finally, the background color of the main content areas—including article body, interactive content, feature body, comment body, and the element with the data attribute “body”—is set to the dark background color on both platforms. This applies to feature, standard, and comment article containers.Here is the rewritten text in fluent, natural English:
On iOS devices, when viewing feature articles, the first letter of a paragraph that comes right after an element atom (or an element atom followed by a sign-in gate) should be styled in a special way. This applies to several sections of the article, including the main body, interactive content areas, the feature body, the data body, and the comment section.
The same styling also applies to standard articles and comment articles on iOS. In these cases, the first letter of a paragraph after an element atom (or an element atom with a sign-in gate) is styled in the article body, interactive content, feature body, data body, and comment body sections.
On Android devices, this styling is applied to the feature article container as well.Here’s the rewritten version in fluent, natural English:
On Android devices, when viewing feature, standard, or comment articles, the first letter of the first paragraph after an element atom should be styled in a special way. This applies whether the paragraph comes right after the element atom, or after a sign-in gate that follows the element atom. The same rule applies to paragraphs in the article body, interactive content sections, feature body, comment body, and any section with the data attribute “data-gu-name=body”..element-atom + p:first-letter,
body.android #comment-article-container #comment-body .element-atom + .sign-in-gate + p:first-letter,
body.android #comment-article-container #comment-body .element-atom + #sign-in-gate + p:first-letter {
color: var(–new-pillar-colour, #ffffff);
}
body.ios.garnett–type-comment #comment-article-container .furniture-wrapper .standfirst,
body.android.garnett–type-comment #comment-article-container .furniture-wrapper .standfirst {
padding-top: 24px;
margin-top: 0;
}
.prose h2 {
font-size: 24px;
}
body.ios #feature-article-container #caption-button,
body.ios #standard-article-container #caption-button,
body.ios #comment-article-container #caption-button {
padding: 6px 5px 0;
}
body.android #feature-article-container #caption-button,
body.android #standard-article-container #caption-button,
body.android #comment-article-container #caption-button {
padding: 4px 4px 0;
}
@media (prefers-color-scheme: dark) {
:root:root:not([data-color-scheme=light]) {
–follow-text: #dcdcdc;
–follow-icon-fill: var(–darkmode-pillar);
–standfirst-text: #dcdcdc;
–standfirst-link-text: var(–darkmode-pillar);
–standfirst-link-border: var(–darkmode-pillar);
–byline: var(–darkmode-pillar);
}
}
body.ios,
body.android {
background-color: #fff;
}
body.ios #feature-article-container .furniture-wrapper .content__labels,
body.ios #standard-article-container .furniture-wrapper .content__labels,
body.ios #comment-article-container .furniture-wrapper .content__labels,
body.android #feature-article-container .furniture-wrapper .content__labels,
body.android #standard-article-container .furniture-wrapper .content__labels,
body.android #comment-article-container .furniture-wrapper .content__labels,
body.ios #feature-article-container .furniture-wrapper h1.headline,
body.ios #standard-article-container .furniture-wrapper h1.headline,
body.ios #comment-article-container .furniture-wrapper h1.headline,
body.android #feature-article-container .furniture-wrapper h1.headline,
body.android #standard-article-container .furniture-wrapper h1.headline,
body.android #comment-article-container .furniture-wrapper h1.headline {
font-weight: 700;
}
.article .article__body h2,
article.content–interactive [data-gu-name=body] h2 {
font-weight: 200;
}
.article .article__body h2:has(strong),
article.content–interactive [data-gu-name=body] h2:has(strong) {
font-weight: 700;
}
.content__meta-container_dcr > div > gu-island {
display: block !important;
}
—
Three paragraphs, from three different hotel reviews. Can you tell which, if any, were AI-generated?
“The hotel is in a great location for everything. Lots of places to eat and drink. The hotel itself is always abuzz. The tavern located on the ground floor is definitely a must. Food, service, prices and atmosphere were great.”
“A good hotel, though the room had the proportions of a well-appointed lift. Slept well, shower was excellent, staff were friendly. Breakfast was busy but competent. Would return, though probably not with a very large suitcase.”
“Excellent base for a London trip. The room was quiet, the bed comfortable, and everything worked exactly as it should. Staff were helpful without hovering. A smooth, unfussy stay from start to finish.”
How do you think you did? Most people, says Claire Hardaker, a professor of forensic linguistics at the University of Lancaster, get this kind of judgment right only about 60% of the time. Her online test, Bot or Not, asks users to identify the fakes in a series of 15 reviews. This middling success rate might surprise those who think they can spot AI writing from a distance. When doubts were raised in May about the authenticity of a prizewinning short story by Jamir Nazir, social media users were quick to condemn it. “If you know, you know,” one commented. Nazir later told The Atlantic that he didn’t use AI.
Hardaker says her respondents tend to rely on a few quick rules of thumb to identify AI language, including clichés and the use of dashes. The “rule of three,” where words or phrases are arranged in a satisfying trio, is also thought to be a giveaway. “People have learned very simpleThere’s a problem, though: these “tells” are also common in human writing—after all, that’s what the large language models (LLMs) that produce them were trained on. “You could go back to Charles Dickens and say he used AI, because he used the em dash too.” And speakers have known about the rule of three ever since Julius Caesar said Veni, vidi, vici. In our hotel review examples, only the first one was real. Did you notice?
Maybe because it’s so hard to know for sure, suspicion has become the norm. In the literary world, writers now face accusations of using AI, with varying levels of evidence. A debut horror novel, Shy Girl, was pulled by publisher Hachette after online rumors that the author had relied on AI—something she denies. Steven Rosenbaum’s book The Future of Truth, a serious look at “how AI reshapes reality,” was found to contain many made-up quotes, which the author admitted in an apology.
Media organizations, including the Guardian, are getting more and more complaints about text that people think is AI-generated. These complaints include gut feelings about certain phrases, as well as comments about typos and grammar mistakes. In one case, the word “after” was accidentally repeated in a sentence. “I can’t imagine a human editor or proofreader missing something like this,” one reader wrote, showing a touching faith in our editing skills.
The problem is that not only does AI learn from human writing, but humans are also influenced by AI’s style. This back-and-forth creates a kind of linguistic hall of mirrors. Unless an author admits it, it’s hard to say for sure whether a piece of writing is AI or not. That uncertainty leads to paranoia.
And if you’re thinking of using a commercial tool to tell human writing from machine writing, that comes with its own uncertainty, says Hardaker. “Given that some of us naturally write in a way that would be seen as AI-like”—she mentions neurodivergent people, for example—“that will be flagged as AI. And you can tweak AI output to make it seem more human. If you put that kind of content into an AI detector, you’ll get weird results.” As someone who has been an expert witness in court, she’s “very skeptical” about how well these tools work.
The newly popular detector Pangram, which claims a false positive rate of about 1 in 10,000, has been shown in independent tests to be very good at spotting AI writing, even when it’s been run through a “humanizer” app to hide its origin. But questions remain. I was able to fool it on my first try (see the screenshot below) by using a bombastic style that could be typical of AI, but could also come from someone who naturally writes that way—or, more to the point, a writer who has been heavily exposed to the output of the LLMs that power ChatGPT, Claude, and Gemini. That, increasingly, is all of us.
View image in fullscreen Photograph: David Shariatmadari
Huge amounts of AI writing are now published every day—from ad copy to academic abstracts and fiction. At the same time, it plays a bigger role in our lives through auto-generated email suggestions, “AI overview” search results, and responses to our chatbot questions. With this level of exposure, it’s no longer a question of whether AI is changing language—both how we speak and how we write. The question is how. And should we resist it, or embrace it?
We’ve known for a while that LLMs tend to produce text that is slightly different from human writing, on average. This often only becomes clear when you look at large amounts of material. One sharp-eyed researcher linked the sudden popularity of the word “delve” to LLMs back in 2024, after searching a database of scientific papers. Other “focal words” that AIs have tended to overuse include “showcase,” “boast,” “underscore,” “garner,” “align,” and “surpa”The words “ss” and “intricate” might appear in any piece of writing without raising suspicion. But here’s another twist: some researchers believe the “delve” trend isn’t caused by the AI models themselves, but by the people who evaluate and guide them through a process called “reinforcement learning with human feedback.” For workers who are “underpaid, stressed, and under time pressure,” it seems that “certain words are treated as a sign of quality,” so the model is unintentionally trained to use them more often. In other words, “delve” might have become so popular simply because it doesn’t sound like a word an AI would use. (One theory suggested it appeared more often because it was common in Nigerian English, where many of these workers lived, but the data doesn’t support that.)
There are other patterns we can spot: large language models love nouns, but they use pronouns less than humans do. This might be because they don’t talk about themselves or others as much as we social creatures do. They prefer attributive adjectives (like “the uncomfortable chair”) over predicative ones (like “the chair was uncomfortable”), possibly because they like to pack information into small, dense chunks, while we tend to spread things out. Different models have clear quirks—you might even call them “dialects.” For example, Gemini often says “here’s a breakdown,” while Deepseek frequently replies with a cheerful “Certainly!” When asked to edit formal English from around the world, AI tends to flatten and standardize it toward an Anglo-American norm, a process researchers call “cultural ghosting.” So, a perfectly acceptable request in Indian professional English like “Kindly do the needful & revert back at the earliest” gets “corrected” to “Please complete the task & respond promptly.”
Evidence that AI language has leaked into the “real” world, changing how humans speak when AIs aren’t around, is now piling up. One study analyzed thousands of unscripted conversations and found that words like “delve” and “boast” spiked after ChatGPT was released. Another study showed that the frequency of “delve” in academic abstracts actually dropped after it was called out on social media, suggesting that AI’s influence might play out in complex ways.
Does any of this matter? Language is always changing—words go in and out of style, and new technology has always been a driver of that. But AI seems to cause especially high levels of anxiety. Why? “I think it scares people because it feels like an intrusion into sentience, like it’s becoming the new human,” says Hardaker. Since 2023, she has expanded the Bot or Not project to include speech and music, and she’s noticed how strongly people react when a song they enjoyed turns out to have been composed and performed by a machine.
Gary Shteyngart, a novelist who teaches creative writing at Columbia University, saw a similar intensity of feeling among his students about the idea of AI literature. “When one of my graduate students said, ‘As an experiment, I’m going to write part of this piece with AI,’ the other students got so angry that they wrote me letters saying how awful this was.”
“There’s an unspoken agreement between writer and reader that the work you’re getting is created by a human being, and I think it felt like a violation of that,” he says. “Reading literary fiction is like an incredible Vulcan mind meld with another person, entering someone else’s consciousness. With AI, I’m entering a simulation of another person’s consciousness, one step removed, or many steps removed. How sad is that in comparison?”
For Hardaker, “I guess it touches on what we think makes us special, what makes us valuable and unique.” At the same time, the music-generation model she uses…”It’s produced some absolute bangers. I listen to them, unironically, in my car, and I really enjoy them.” Could the same thing happen with literature? Will a machine-written novel one day rank among the 100 greatest of all time? Peter Stockwell, a professor of literary linguistics at the University of Nottingham, thinks AI can handle the basics but can’t reach the highest levels. “If you want something very familiar, very average, and completely functional, it’s amazingly good at that.”
He explains that language can be seen as a series of layers, starting with words, then phrases, clauses, compound sentences, and finally narrative structure. “AI is really good at the lower levels. It’s learned a lot of our sentence structures, so everything looks well-formed and grammatical. But the higher you go, the worse it gets.” Getting the arc of a story right is especially hard for AI.
“If you ask AI to write a story, it can do a decent job of having a sequence of events and something happening at the end. But it wouldn’t be a very compelling narrative,” he continues. “Nothing surprising or interesting would happen. And if something surprising does happen, it usually looks like a mistake, not a brilliant twist.”
The secret to great writing remains a mystery—even to the academics who study it. “Linguists don’t really understand how language works at its higher levels,” like discourse, storytelling, and enchantment. “We can’t build a machine to do something when we don’t know how it works.” We do have some idea of what it might come down to—our fundamentally social nature and the fact that we are “wetware”: human flesh with spikes of adrenaline, rushes of dopamine, and a craving for social contact, all of which show up in language’s structure and how we use it.
Would I work with an LLM? Of course! Why not?
—Jeannette Winterson
Stockwell explains that there are two main models in linguistics. One sees the brain as a computer that parses grammar and computes meaning. The other sees it as fundamentally embodied, reflected in language by how we understand things through “seeing” or think of “up” (where our head is) as good—we get “high” and feel “low.” “One key thing is that current AIs don’t have a body. They don’t exist in the world, so they don’t know what it feels like to be in the world as a human.”
For Shteyngart, feeling is essential: “Today is the first warm day in New York. If I were to start writing a novel, I think it would be warmer. I think I would filter what I know through the warmth of the day. If I ate a really wonderful lunch and sat down to write, there would be more sensuousness in my writing.”
“The love of the body and its encounters with the physical world drives some of the best literature. So I almost feel sorry for these LLMs, as I talk about them, because they’re stuck in some horrible machine in the Bay Area, and they just don’t know how wonderful life is.”
One big fear about the widespread use of LLMs is that they act as a flattening force—smoothing out the variety and quirks of human language into a kind of bland goo. That’s a valid concern, as far as it goes, but it’s not new. People have long worried about the homogenizing effects of American film and TV on accents and vocabulary. There are also language subgenres—like political euphemism, customer-service chatter, and therapy-speak—that have spread further than many would like. But the key point is that their influence often sparks a backlash—and there’s no reason to think it’ll be different this time.
In fact, our capacity for innovation might ultimately be what truly sets humans apart.Writing—especially literary writing—from AI. “The whole point of an LLM is that it’s trained on existing language. So it’s always looking backward,” says Stockwell. “I could ask an AI to ‘write me a short story in the style of Virginia Woolf’ and it would do a decent job. But you can’t say, ‘write me a story in the unique style of the next great, serious literary innovator.’ It couldn’t possibly do that.”
That’s because, once again, it lacks the social environment and the body that create characteristically human motivations. “Why does someone do something new in an art form like literary writing? It could be out of annoyance or irritation with what came before. Or because someone sees things differently than the average person, or sometimes just because people are restless and want to do something different, or a little bit crazy or isolated.”
There are plenty of examples from history, says Stockwell: “After the bureaucracy and uniformity of World War I, you get this sudden, huge, opposing artistic movement with the rise of surrealism and Dada. Similarly, after the austerity of World War II, you get the psychedelic movement, and art and literature change again, quite radically. So there always seems to be that kind of pushing back against the norms. It’s hard to imagine how you would program an AI to do that, because AI works on an existing large body of material. It’s the embodiment of the conservative status quo.”
Originality is so important for novelist Jennifer Egan that she’s completely cut herself off from the technology. “I feel a danger of infection, to use a kind of loaded metaphor,” she tells me. “I know they stole some of my stuff, and there’s nothing I can do about that, but I’m not giving them one more word voluntarily.” Anthropic used pirated copies of books, including Egan’s novels, to train its chatbot Claude; most LLMs use language from individual queries as additional training data. She sounds frustrated: “I don’t want to take part in this kind of language spam they’re offering.”
Her zero-tolerance policy doesn’t stop her from feeling paranoid. “I’ve been told a couple of stylistic things that are signs of AI, and they happen to be things I like. For example, I love em dashes, but now I find myself questioning every single one way more than I used to. I’ve also noticed that I tend to use groups of three. So I find myself questioning those too. I don’t mind that, actually, because the whole point is to not write something that anyone else could have.”
What advice would she give a younger writer now swimming in this environment? Should they practice their own kind of AI hygiene? “I’m going to sound like the totally generic boomer that AI could probably have written, and my advice is: stay the hell away. I mean, OK, use it to write emails. Even use it to get research ideas. But if you want to be a writer: learn to write. Come on. I would really question why you’d feel the urge to use it.”
Not everyone is so strict. Jeannette Winterson, who has written a lot about AI and art, tells me: “Every writer can make their own choice. Humans are tool-using animals. That has been our success story. Right now, all AI, including generative AI, is a tool. Would I work with an LLM? Of course! Why not?”
But she warns against the idea that AI’s language skills mean it can match or surpass human expression. “Beyond the basics, meaning becomes a series of inner realities, and language is great at conveying those inner realities. Machines don’t share our reality, especially because they don’t have a limbic system. Humans can’t have a thought without a feeling … literature is brilliant at revealing these layers.”
As I paste her quotes into a Google doc full of my notes, I notice the built-in AI making a suggestion. It asks if I want to change Winterson’s words to more closely “match the style” of the existing material.The goal is to soften the rough edges of one of the English language’s most unique writers. With almost superstitious urgency, I brush off the prompt.
Frequently Asked Questions
Here is a list of frequently asked questions about how AI is changing the way we use language
Beginner Questions
1 What does it mean when people say AI is changing language
It means AI tools are now helping us write emails summarize articles and even create stories This changes how we think about grammar tone and creativity because a machine is handling part of the communication
2 Is AI going to make us forget how to write properly
Not necessarily Think of it like a calculator it helps with math but you still need to know the basics to check if the answer is right You still need to know good grammar to guide the AI and catch its mistakes
3 Can AI understand sarcasm or jokes
Not really AI can recognize patterns of sarcastic language but it doesnt get the humor or emotion behind it It often takes things literally which is why AIgenerated jokes can feel flat or weird
4 How is AI helping people learn new languages
Apps like Duolingo and Google Translate use AI to give instant feedback on pronunciation and grammar They can also simulate conversations which helps you practice without feeling embarrassed
5 Is it cheating if I use AI to write an email or a school essay
It depends on the rules Using AI to fix typos or rephrase a confusing sentence is usually fine Having AI write the entire thing for you without adding your own thoughts is often considered cheating or plagiarism
Intermediate Advanced Questions
6 How does AI learn to understand human language
Through a process called training Engineers feed the AI billions of sentences from books websites and transcripts The AI looks for patterns and builds a giant statistical map It doesnt know meaning it just predicts the most likely next word
7 Why does AI sometimes make up facts or hallucinate
Because AI is designed to sound confident not to be truthful If it doesnt have the right data in its training it will still generate a plausiblesounding answer Its like a student who guesses an answer to avoid saying I dont know
8 What is prompt engineering
Its the skill of writing the right instructions to get a good result from AI
