What changed for shared links
For most of the history of share buttons, the question behind a shared link was whether the preview looked right. The title, the description and the image were the whole story, and Open Graph tags in WordPress settled all three. Since generated images and text became cheap, feeds have filled with material nobody wrote or photographed, and readers have started asking a second question when a card appears: who made this, and was it a person?
Platforms have responded with labels, disclosure toggles and systems that read provenance data inside image files. None of these is perfect, and none of them is under your control once a reader shares your page. What you can control is what your own pages say about themselves, which is the subject of this section.
Four things a publisher can do today
Decide and publish a disclosure policy
Write down when you use generative tools (research, drafts, illustrations, translation) and how you say so. Put the policy on your about page and follow it on every post. Consistency matters more than the exact wording.
Label generated images at the source
If an illustration was generated, say so in the caption and keep the file's provenance data intact where your tools allow it. Platforms that read that data can then label the image correctly instead of guessing.
Name a real author
Every article gets a byline that links to an author page, and the page is backed by structured data. The guide to marking up a human author with byline schema covers the markup.
Carry authorship into the preview
Add author tags to your page head so networks that support them can show who wrote the piece next to the card. See author meta and fediverse:creator tags.
Why this matters for reach, not just ethics
Labels change how posts travel. A photograph that a platform tags as generated may be shown with a notice, ranked differently in some surfaces, or questioned in replies. The inverse also happens: honest photographs get flagged because an editing step left metadata that looks like generation. Knowing how platforms decide when to label AI content helps you avoid both mistakes.
Search engines take a different line. Google's public guidance says it rewards helpful content however it is produced, while its spam policies target pages mass-produced to manipulate rankings. The practical reading for publishers is the same as for social feeds: be accountable for what you publish, and make that accountability visible.
Text is harder than images
There is no reliable provenance standard for text in wide use. Automated detectors that claim to tell human writing from generated writing produce false positives, especially on short, formal or edited text, and a share card offers them even less to work with. That is why the signals that work for articles are editorial ones: a named author, a visible reviewed date, a disclosure line where tools were used, and a publication history readers can check.
| Signal | Works for | Survives a share? |
|---|---|---|
| Caption or disclosure line | Images and text | Only if the reader opens the page |
| Content Credentials in the file | Images and video | Sometimes; many uploads strip it |
| Author meta and fediverse:creator | Articles | Yes, on networks that read them |
| Byline schema | Articles | Read by search engines, not feeds |
| Platform label | Anything uploaded natively | Set by the platform, not by you |
What not to do
Do not claim that an image is unedited when it is not, and do not strip provenance data to avoid a label. Both approaches tend to surface later in replies or corrections, and the damage to trust costs more than the label ever would. Do not add AI disclosure badges to share buttons or preview images either; a card is too small for nuance, and the explanation belongs on the page.
AI-generated content on social media: a worked example
Take a site that publishes 40 posts a month, of which 12 carry a generated illustration and 6 used a writing assistant for a first draft. The counts are invented for this example, but the pattern is typical of a small publication. Under the policy above, the 12 illustrations get a caption naming the tool's role, the 6 drafted posts get a one-line disclosure in the description, and all 40 carry an author meta tag and a fediverse:creator tag for the writer.
Sharing 10 of those posts to four networks gives 40 cards to inspect. In the example, 38 of the 40 cards show the intended title and description, so the disclosure lines written into descriptions reach card readers on every network that shows descriptions. The Mastodon cards show the author line on 8 of 10 posts; the 2 that do not belong to a guest author who has not approved the domain. None of the 40 cards shows a Content Credentials badge, because every network re-encoded the preview image. Page-level signals travelled and file-level signals did not, which is the expected result rather than a fault.
Pitfalls
- Disclosing in a footer widget only. The share card never carries it, and a reader who arrives from a feed may leave before scrolling that far.
- A policy page that is stricter than practice. Readers compare the two, and the gap does more harm than a looser policy that is followed.
- Assuming the featured image is the share image. Many themes and SEO plugins pick a different crop for og:image, so the caption and the card can describe different pictures.
- Treating a detector score as a publishing decision. Scores on short text are unstable, and a false positive on your own writing is common enough that it should not drive edits.
- Changing bylines to a house name to avoid accusations. That removes the one signal readers use most to decide whether a person stands behind a post.
What to record
Keep a short share test log for each post that carries generated media or a disclosure line. One row per network is enough. The log tells you what a reader actually saw at the time, which matters when a label or a complaint arrives weeks later and the cache has moved on.
| Page | Network | Time (UTC) | Preview showed | File carried |
|---|---|---|---|---|
| /2026/09/harbour-report/ | 2026-09-22 09:14 | Title, description with disclosure line, 1200 by 630 crop | No manifest after re-encode | |
| /2026/09/harbour-report/ | Mastodon | 2026-09-22 09:20 | Card with author line for @ana@example.social | No manifest |
| /2026/09/harbour-report/ | 2026-09-22 09:31 | Title and image; description trimmed at about 140 characters | No manifest | |
| /2026/09/harbour-report/ | X | 2026-09-22 09:40 | Summary card with large image, no description | No manifest |
Re-run the test after a theme change, an SEO plugin update or a CDN change, because each of those can alter which image or description a network fetches. The share preview integrity protocol sets out the full routine, and the provenance hub lists the guides for each signal.
Sources
- Meta, Our Approach to Labeling AI-Generated Content and Manipulated Media (April 2024)
- Google Search Central: guidance on using generative AI content
- Regulation (EU) 2024/1689, the AI Act (EUR-Lex)
Questions
Do I have to label AI-generated images on my site?
Rules depend on where you publish and what the image shows. Several platforms require disclosure for realistic generated media, and the EU AI Act adds transparency duties for some deepfakes. A clear caption is the simplest habit.
Will labelling AI assistance hurt my reach?
Not by itself. Undisclosed generated media that is later flagged tends to do more harm than an honest disclosure line.
Can a share card show that a human wrote the article?
Partly. Author tags let some networks show the writer beside the preview, and byline schema tells search engines who the author is.
Should AI disclosure go in the share card image?
No. Print nothing onto the 1200 by 630 image. Put the line in the description or the first sentence, which networks show as text, and keep the fuller note on the page.
How often should I re-test how my posts look when shared?
After every theme, SEO plugin, CDN or image optimiser change, and once a quarter regardless. Networks cache cards, so test a fresh address each time.
Hannah Voss, Editor. Checks every guide against a working WordPress install and the networks' current documentation. Last reviewed September 2026.
Related reading
- Authorship tags for shared links: author, article:author and fediverse:creatorAdd authorship signals to link previews: the author meta tag, article:author, twitter:creator and fediverse:creator, with WordPress code and testing steps.6 min read
- How platforms label AI content, and what triggers the labelHow major platforms decide to label AI-generated or altered media: creator disclosure, provenance metadata, detection, and how publishers avoid mistakes.6 min read
- Content Credentials on shared images: what survives a shareWhat C2PA Content Credentials are, why WordPress image sizes and social uploads often strip them, and how to keep provenance on the images you share.6 min read
- Adding Open Graph tags in WordPressAdd Open Graph tags to WordPress with an SEO plugin or theme code: the four required tags, image tags, per-post overrides, and how to avoid duplicate tags.7 min read