Photograph the problem you fix and write the answer underneath

What Google's patent filing describes

In a patent published this year, Google's own search team describes how a photo question gets answered. The system finds the stored photo that most closely matches the customer's photo. Then it builds the answer out of the words printed next to that photo, on whichever page holds it.

The document is US20260111481A1. It carries a priority date of May 2023 and was published on 23 April 2026. Andy Chadwick, an independent search analyst, found it and wrote it up in July 2026.

Going by the claim, four things happen in order. A customer sends a photo and a question, asked together. The system reads that photo and compares it against photos it already holds, then picks the closest match. From there it goes to the page that holds the matching photo and pulls a block of text off it. A language model writes the answer out of that text, and the photo gets shown alongside it as the source.

Read that sequence twice, because the order is what matters. A page gets found through its photo, and then the customer hears back whatever sentence happens to sit beside that photo. Both of those things live on somebody's website. Today, for most service businesses, it's probably somebody else's.

I put the lead image from one dental homepage into a reverse search this week. The same picture turned up on clinic site after clinic site, all of them built on one template.

Four caveats travel with all of that, and they matter. A patent application is an idea a company has asked to protect. Google files thousands every year, and plenty of them never ship the way they were written. Nothing here has been confirmed as live behavior in a Google product today.

The filing also covers one route through a much larger system. Real answers very likely blend several signals at once, and this claim spells out one of them. It applies to one kind of question too: a photo and a question, asked together. A typed prompt in ChatGPT, or a typed search that returns a Google AI Overview, runs through a different pipeline.

One more. The filing says the answer text can come from around the matching photo. It stops short of naming how close. So keeping your answer near the photo is sound practice, and the filing itself promises nothing about how near.

Nearly 20 billion camera searches a month

Google says Google Lens handles nearly 20 billion visual searches every month. By Google's own count, about 20% of those, close to 4 billion a month, are shopping-related. Both numbers come from Google's own reporting, and no outside auditor has checked them.

Every month, people point a phone camera at something they can't name and ask what they're looking at. Some of those photos show a thing your business fixes. A cracked filling. A rusted valve behind a water heater.

That customer is holding a picture and a worry, with no name for the part, and whatever comes back first shapes who they end up calling. If you run an HVAC company, you've answered that same question on the phone a hundred times. The answer is already in your head, and writing it under a photo takes about a minute.

A photo you took yourself, in your own treatment room or on your own job, exists on exactly one page on the internet. That page is yours. When the sequence in the filing runs, the page it lands on is the one with the matching photo, and the words it reads out loud are the words you put there.

The words under the photo are what a customer hears back

The caption is the second half of the sequence. Myriam Jessier made the same practical point in Search Engine Land on 27 August 2026. She tells operators to shoot their own photos, and to put the fact a buyer needs right under the image. Her worked example is a plumber who posts photos of leaky faucets with plain text beside each one, explaining the problem and how the service fixes it.

Chadwick puts it in one line. "Your image is the ticket into the answer; your text is the plus-one," he wrote in July.

An HVAC company can photograph the failed capacitor it pulled out of a unit on a real job. Underneath goes one line: what the part does, and what a replacement runs. A vet clinic already has a folder of paw and ear photos on a phone somewhere. Most of them have never been posted anywhere, and each one could carry the sentence an owner types at 10pm. For a dental practice, a worn crown works well, described generically and with the patient's consent.

A medspa is the harder case. Keep it light: photograph the machine and the room it sits in, then write what the treatment is for and roughly what it costs. An in-home care agency can do the same with the equipment it sets up in a house.

What goes in the caption is the sentence you already say out loud on the phone. Name the thing in the words a customer would use, then say what it costs to put right. Somebody holding a photo of a cracked filling wants the name of the problem and a price range. The line about caring deeply can go somewhere else on the page.

Three steps for this week

Chadwick sorts his advice by budget, from a full photography program down to a free check you can run yourself. Taken cheapest first, that gives three steps, with the time each one costs.

1. Pick your best-earning service page and reverse-search its main photo in Google Lens. Count how many other sites carry the same image. A high count means the photo came from a supplier feed or a photo library, and it lives on hundreds of pages that belong to other people. Ten minutes.
2. Ask whoever built your site to confirm that image crawlers aren't blocked in your robots.txt file, the small text file that tells search bots which parts of your site they may read. Five minutes of somebody else's time.
3. Swap one decorative banner photo on that page for a single original photo, taken at the spot where the question actually gets answered. Write the answer directly under it, in the words a customer would use. An hour, most of it spent walking around with your phone.

Step 3 is the one that pays. The photo is what gives your page a chance of being matched at all. Then the caption does the answering, which is the part a customer actually reads. Both of those belong to you the moment you take the picture.

None of this rests on the patent shipping. An original photo of your own work, with a plain answer under it, does the same job on your website today that it would do inside the sequence Google described. It tells a customer what the thing is and what it costs to fix, in one screen, without a form in the way. That is worth an hour whether or not the filing ever runs in a Google product.