Alt text describes the picture to somebody who cannot see it. That is the whole specification, it was written for accessibility rather than for search, and doing it properly is also the version that works in image search. Hebden Hats makes and sells hats in Hebden Bridge, rewrote 60 alt attributes in February 2026, and image search sent 30 visits a month and now sends 240.
The rewrite is not clever. hat1.jpg with alt text hats hebden bridge milliner became a file called grey-felt-wide-brim.jpg with alt text wide brimmed grey felt hat with a petersham band, and 59 others were rewritten the same way. A person hearing the second one knows what is in the picture. A person hearing the first one has heard a shop's marketing read aloud.
Which images get alt text and which do not
If the picture is decoration, the alt attribute is empty and that is correct. An empty alt on a background flourish tells a screen reader to skip it, which is the right outcome; leaving the attribute off entirely makes the same reader announce a filename. Hebden had 22 decorative images with alt text describing them, and all 22 were removed rather than improved.
Decisions
How Hebden Hats sorted images between empty alt, alt text, and captions
Every image belongs in one of three lanes depending on whether it decorates, depicts a product, or adds unseen history.
- Empty alt attribute (decoration)2
- Skip cleanlyBackground flourishes
- Removed22 decorative images cleared of alt text
- Descriptive alt text2
- Screen reader58 product photographs described plainly
- Omitting alt so readers announce raw filenames
- Visible page caption2
- Everybody2 photographs needing non-obvious context
- Context1940s block detail placed in visible caption
Alt text serves visitors who cannot see the image, while captions carry facts that even sighted readers would miss.
An image needs a caption when the point is not obvious from the page. Alt text is not read by a sighted reader and a caption is read by everybody, so the fact that the grey felt is a 1940s block belongs in a caption rather than in the alt attribute. Two of Hebden's product photographs now carry captions and the other 58 do not.
Where the ranking part actually comes from
The page around the image decides most of what the image is taken to be about. A photograph of a hat on a page about petersham ribbon is understood as a ribbon photograph; the same file on a product page is understood as that product. Alt text and filenames are the smallest inputs and the page is the largest. Moving three photographs onto the pages they belonged on accounts for about 96 of the extra visits a month, the 60 rewritten attributes for about 74, and renaming the files for about 12, which is the honest ordering of the three.
A filename is a weak signal and it costs nothing to get right. Weak is the accurate word: nobody at Hebden can point at a single visit caused by a filename. It takes four seconds at upload, it helps a little, and treating it as the important half is how people end up with files called grey-felt-hat-hebden-bridge-milliner-handmade-hats-uk.jpg.
Where this stalls: alt text written for a machine
Stuffing keywords into alt text is read by a screen reader out loud, which is the fact that ends the argument. Somebody using a screen reader on Hebden's old site heard hats hebden bridge milliner sixty times in one page visit, which is both useless and unpleasant, and it did nothing for search either.
Where it stalls
The spreadsheet loop that turns alt text into keyword noise
Driving missing-alt counts to zero in bulk produces repetitive labels that punish screen-reader users and fail in search.
- 01Audit flags missing alt count
- 02Half of 60 written from a spreadsheet
- 03Same shop slogan repeated aloud
- 04Zero ranking gain and poor usability
- 01Audit flags missing alt countA countable metric that feels satisfying to drive down to zero.
- 02Half of 60 written from a spreadsheetFilled out in one sitting without looking at individual photographs.
- 03Same shop slogan repeated aloudScreen readers announce hats hebden bridge milliner sixty times.
- 04Zero ranking gain and poor usabilityLeaves real page-placement and description work undone.
Describing the actual object in the frame takes no longer than repeating a keyword string, and helps both people and search.
The stall in its ordinary form is an audit tool reporting a count of missing alt attributes. It is countable, it is satisfying to drive to zero, and it produces sixty attributes written by somebody filling in a form rather than describing photographs. Half of Hebden's original 60 were written in one sitting from a spreadsheet, and every one of them said roughly the same thing.
The part with the largest cost attached
Compress before uploading, because nothing on the page costs more to load. Hebden's product photographs were 3.4 megabytes each straight from a camera, and a page with four of them was 14 megabytes; resized to 1600 pixels and saved as WebP they are about 180 kilobytes each. That is a change to how fast it loads, and what that is worth in search more than to image search, and it is the image decision with a number attached.
Cost and measurement
File weight cut before upload, and hidden visits tracked in Search Console
Resizing camera originals fixes page load cost, while the image search tab uncovers traffic buried inside combined totals.
- 3.4Original camera file (megabytes)Four files put 14 megabytes on a single page before resizing.
- 180Resized WebP file (kilobytes)Saved at 1600 pixels; Screaming Frog still flagged 74 files over 100 KB.
- 30Monthly image visits beforeVisible only after separating the image tab in Search Console.
- 240Monthly image visits after210 extra visits that disappear inside the default web total.
Compression is the image choice with the largest page-speed cost, and Search Console's image tab is where the lift appears.
Search Console separates image search from web search in its performance report, free, and it replaced the keyword data Google removed from analytics in 2011. The image tab is the only place the 30 and the 240 exist; in the default view they are folded into a total where 210 extra visits are invisible.
Screaming Frog SEO Spider lists every image over a size threshold and every missing or duplicated alt attribute, free up to a few hundred addresses, and it replaced checking each page in a browser. Its useful export here is images over 100 kilobytes, which at Hebden was 74 files.
What the page containing the photographs is for is what separates a page that ranks from one that exists, and the alt text is written under the same rule as an outline a reader and a crawler both use: say the specific thing. Both are inside what a single page controls about its own ranking.