Can AI actually see the photos on your site?
Less than you would hope, and never on its own. The picture of your finished bathroom, the before and after, the shot of the team outside the shop: all of it reaches a machine as a file name and whatever words you have put near it.
Google is the only one of these companies that writes down how it handles images. Its guidance for site owners says that it uses the written description attached to a picture, along with software that looks at the picture itself and the words on the page, to work out what the image shows.1 That written description is usually called alt text. It sits behind each image in the page, most visitors never see it, and it is read aloud to people using screen readers.
So Google does look at the photograph. Read the rest of that page, though, and the emphasis is unmistakable. Google says it takes information about the subject of an image from the content of the page, including captions and image titles, and it asks you to make sure images sit near text that is relevant to them.1 The picture is evidence. The words are the argument.
The assistants say less. OpenAI publishes a page describing the three automated readers it sends out to websites, including OAI-SearchBot, the one it uses to surface websites in ChatGPT's search features. None of the three descriptions mention photographs at all.2 That is not proof that pictures are ignored, and nobody outside these companies can prove it either way. It is a fair signal of where your effort is repaid. Google's own rule for appearing in its AI answers points the same way: a page has to be indexed and eligible to show a line or two of its text in ordinary search.3 Text is the currency.
The file name
Google calls this a very light clue. IMG_9520.PNG is not a clue at all.
The written description
The short line of text behind the image. This is the one that gets skipped.
The caption
Visible text under the photo. Google names captions and image titles specifically.
The words around it
A photo on a page about the wrong thing is a photo about the wrong thing.
What we found on 470 UK service business websites
Nobody had measured this for UK service businesses, so we did it on 30 August 2026. We took 629 dentists, GP practices, vets, hairdressers, beauty salons and garages that record a website in OpenStreetMap across London, Manchester, Birmingham, Glasgow and Cardiff, and read each homepage once the way an automated visitor would.4 470 of the 629 loaded for us. Between them they carried 10,533 images, a median of 12 per homepage.
27.3% of those images carried no written description at all. Some of that is deliberate and correct: 20.8% had the description left deliberately blank, which is the proper way to mark a decorative swirl or a divider so a screen reader skips it. The other 6.5% simply had nothing there.
The number that matters more is what the descriptions said when they existed. We counted a description as useful only if it ran to three words or more and named the trade, the town or the business itself. Just 17.2% of all the images cleared that bar, and 42.4% of the sites with any images at all did not have a single one.
What sits in those description fields, when somebody has filled them in, is the part worth seeing. Real examples from the sites we read, with the businesses left unnamed: image.jpg. IMG_9520.PNG. thumb1 banner, repeated on five photographs in a row. A Cardiff barber had done markedly better, numbering five photographs as work examples with the shop name in each, which at least identifies the business and still tells a machine nothing about what is in the picture or which town it was taken in.
None of this is unusual for the web at large. WebAIM checks the top one million home pages every year, and in February 2026 found that 16.2% of all images on them had no written description, and that 10.8% of the descriptions that did exist were generic, repetitive or just a file name.5 Our UK service business sample came out better on the first measure and slightly worse on the second.
So are photos a waste of time?
No, and the reason has nothing to do with AI. Photographs are how a person decides you are real before they ring you, and that job has not changed. What has changed is that a second reader now arrives at the same page and gets almost nothing from them.
There is also a genuine visual search story, and it is worth being precise about how much of it applies to you. Google said in October 2024 that people use Lens for nearly 20 billion visual searches every month, and that 20% of all Lens searches are shopping related.6 That is a global figure across every kind of search, not a UK one and not a services one.
When Google announced its bigger push into visual searching inside AI Mode in September 2025, the examples it chose were home decoration, fashion and finding a particular throw pillow, and it talked about a shopping catalogue of more than 50 billion product listings.7 Local services were not mentioned. Somebody photographing a leaking radiator to find out what it is remains a real and growing behaviour, but the published evidence for visual search is retail evidence, and anyone quoting those 20 billion searches at a plumber should say so.
Do Google Business Profile photos help you rank?
Google does not say they do. Its own page on improving your local ranking names three things and only three: how well you match what was searched for, how close you are to the person searching, and how well known you are.8 Photos appear on that page once, described as a way to show customers what you offer and tell the story of your business. That is a claim about winning the click, not about winning the position.
Google is much more specific elsewhere about what to upload. It asks for at least three exterior photos, at least three interior photos, at least three of the services you provide, and at least three of your management team, and says category-specific photos spotlight the features customers use when deciding to buy.9 Worth doing, and it takes an afternoon.
You have probably been shown a much more exciting number than any of those.
more calls for listings with over 100 photos is the figure quoted in almost every article about business photos. It comes from a study of 45,264 listings covering September 2017 to December 2018, and it shows two things happening together rather than one causing the other.
That study was BrightLocal's, published on 15 July 2019, and to its credit BrightLocal saw the problem itself, noting that businesses with a high number of images "may be businesses that are 'better' at Google My Business in general".10 A dental group with 200 photos also has a marketing budget, a receptionist updating the listing and probably 400 reviews. The photos arrive with all of that, and the study has no way of separating them from it. The same study found UK listings carrying 10 photos against 13 in Canada, which is the more useful number: the bar is low.
What about video?
Almost nobody in this trade has any. Only 11.7% of the 470 UK homepages we read carried a video of any kind, counting both videos hosted on the page and YouTube or Vimeo players embedded in it.4
Whether that matters depends on a study worth reading carefully. Otterly.AI, which sells AI visibility tracking and therefore has an interest in the answer, looked at more than 100 million occasions on which an AI answer quoted a source over 30 days, across ChatGPT, Google's AI Overviews and AI Mode, Perplexity, Copilot and Gemini, and published the results on 2 March 2026. It does not state which countries the searches came from. It found that 5.54% of all those quotations came from social media and video platforms, and that YouTube accounted for 31.8% of the social ones, behind Reddit on 46.4%.11
The interesting part is which videos got quoted. Otterly.AI scored each thing it measured for how closely it moved with being quoted, on a scale running from minus 1 through 0, meaning no connection at all, to 1, meaning the two always move together. View counts, likes and subscriber numbers all landed at roughly zero. The one thing that showed any positive connection, at about 0.3, was how much the video's written description said.
| What was measured | Score out of 1 | What that means |
|---|---|---|
| How much the written description says | 0.3 | A weak but real connection |
| How many likes the video has | -0.02 | No connection |
| How many views the video has | -0.03 | No connection |
| How many subscribers the channel has | -0.03 | No connection |
One more thing before anybody films anything. Google's rules for videos appearing properly in search expect the video to sit on what it calls a watch page, whose main purpose is to show a single video, with its own title and description.12 A thirty second clip dropped halfway down your services page is not that, and will mostly be doing a job for human visitors rather than for search.
What to do about it this week
Nothing here needs a camera crew, and most of it is typing.
Write a real description for every photograph that shows the work. Not the logo, not the decorative swirls, which should be left deliberately blank. Say what is in the picture and where: "new bathroom fitted in a terraced house in Levenshulme" rather than "bathroom1". Whoever built or maintains your site can list every image missing one in a few minutes, and in most website systems it is a box next to the photo when you upload it.
Put words next to your gallery. Google asks for images to sit near relevant text, so a page of twenty photographs with a heading and nothing else is the worst layout you can choose. Two lines under each group, naming the job and the town, changes the whole page.
Upload the photos Google asks for on your Business Profile, since it has published the list: three outside, three inside, three of the work, three of the team. Do not expect it to move you up the results. Expect it to matter when somebody has already found you and is deciding whether to ring.
If you make video, spend the time on the title and description. On the evidence above, that is the part with any measurable connection to being quoted, and it is free.
All of that is a morning's work and none of it needs an agency. What people pay us for is that it stays true, and that the next fifty photographs go up the same way. Frontpage AI is a UK studio that builds and manages websites for independent service businesses, structured so that Google and AI answer engines can read, trust and name them. We work only with UK service businesses, such as trades, clinics, salons and instructors, and charge from £148 per month on a 30-day rolling contract with no setup fee. The way we approach AI search page sets out how, and our piece on what makes an AI name your business covers the signals that matter more than any of this.
What this evidence does not tell you
It does not tell you that fixing your image descriptions will get you recommended. No published study measures that, for any kind of business. What the evidence supports is narrower and still worth acting on: the words attached to your pictures are the only part of them a machine reliably reads, and most UK service businesses have not written any.
Our own check has real limits. It is one morning, five cities and six trades, drawn from businesses recorded in OpenStreetMap, which skews towards places somebody thought worth mapping. 159 of the 629 sites did not load for us at all. We read the homepage only, so a business with beautifully described galleries deeper in the site would look worse here than it deserves. And our test for a useful description is our own: three words or more, naming the trade, the town or the business. A different test would give a different number.
The visual search figures are Google's own and are global, across all categories. The video study is by a company selling AI visibility tools, does not state which countries it covers, and looks only at videos that were already being quoted. The photos and calls figure is seven years old and describes American, Canadian and British listings together.