Authority & Mentions · 9 min read · July 15, 2026
Documenting reference projects correctly so the AI uses them as proof
When a customer asks ChatGPT which joiner in your region builds a solid-wood kitchen or a made-to-measure fitted unit, it's not your gut feeling that decides the answer, but what the AI can read about you. Reference projects are your strongest proof, but only if they're documented machine-readably: with material, dimensions, location, task and result in text form. This guide shows you how.
Why the AI doesn't see your beautiful photos
Most joiners have a gallery on their website: twenty pictures of kitchens, staircases and fitted wardrobes, often without a single word of text. For a person that's impressive. For ChatGPT, Perplexity or Google AI it's nearly invisible. Language models understand images on ordinary websites only to a limited degree and draw their answers predominantly from text. A photo of a walnut kitchen island says nothing to the AI, as long as it's written nowhere that it's a walnut kitchen island.
That's the central error in the trades: we show our skill visually, because we experience it visually. But the machine that advises your future customer reads. It looks for sentences like "made from oiled wild oak, 3.20 meters of work surface, installed in an old building in Regensburg". If these sentences are missing, your best work stays a mute picture without meaning.
The good news: you don't have to build anything new. You only have to describe what you've already built anyway. And because barely any competitor does this, even one cleanly documented project gives you a noticeable lead in the AI answers.
What makes a reference project into proof
Proof is something different from praise. "We build high-quality furniture" is a claim. "In 2025 we made for a dental practice in Augsburg a reception counter from oiled oak veneer with integrated cable management and LED lighting, length 4.10 meters, installed in a single day" is proof. The difference lies in verifiability: concrete numbers, materials, locations and tasks can't be confused with marketing language.
Language models prefer exactly such provable statements, because they can answer a concrete question with them. If someone asks "Who builds practice fit-outs from solid wood in Swabia?", the AI draws on your documented counter project, names you as an example and can even confirm the material and the region. Without these details you stay one of many anonymous joineries.
The rule of thumb for proof is: every sentence should contain a piece of information a customer couldn't invent himself. Material type, wood type, dimensions, installation situation, special feature, timeframe. The more specific, the more credible — for the person and for the machine.
The five building blocks of every project description
So you don't rethink it for every project, use a fixed structure. Five building blocks suffice: First, the task — what did the customer want and what problem was there? Second, the material — which wood type, surface, fittings, board materials? Third, the dimensions and key data — length, height, room situation, quantity. Fourth, the special feature — what was tricky, slanted, heritage-protected or unusual? Fifth, the result — what did the customer get and how long did it take?
An example for a kitchen: "Task: open living kitchen in a terraced house in Fürth, customer wanted seamless fronts and a lot of storage on little floor space. Material: handleless MDF fronts in matte anthracite, worktop from 4-centimeter-thick solid oak, pull-outs with soft-close. Dimensions: kitchen run 3.80 meters plus island 2.10 meters. Special feature: slanted roof surface above the sink, completely adapted to measure. Result: turnkey in six weeks from order."
These five building blocks you can fill in in five minutes per project, often right after installation, when you still have the details in your head. Set up a simple template on your phone and dictate it on the drive back from the site. This way a collection of real, provable references arises without extra effort.
Captions and alt texts: the quiet lever
Your photos stay important, for people. But you can also make them usable for the AI by giving every image a real alt text and a meaningful caption. The alt text is the text description in the background that search engines and AI systems read out. Instead of "IMG_4821.jpg" or "kitchen" you write there: "kitchen island from oiled wild oak with cooktop extractor, made by Tischlerei Muster in Regensburg".
Likewise, under every project image belongs a visible caption in whole sentences. "Fitted wardrobe in the attic, slope cladding from birch multiplex, painted white, width 3.40 meters" is ten times more valuable than no caption at all. These small text blocks add up: a gallery with twenty well-captioned images suddenly delivers the AI twenty usable pieces of evidence instead of twenty mute files.
Make sure the descriptions stay honest and concrete. No keyword stuffing, no invented superlatives. The AI and attentive customers recognize platitudes. A plain, precise sentence about what's really visible in the picture comes across as the most credible.
Integrating customer voices correctly
Reviews are a strong signal for language models, because they provide independent confirmation. But a star rating without text helps the AI little. So ask your customers specifically for a few sentences that name the project: "The joinery built our living-room wall completely to measure with a built-in shelf from ash, including a recess for the fireplace." Such concrete voices link your name to a real achievement.
Most effective is when the same information appears in several places: in your project description, in the customer's Google review and perhaps in an industry directory. When the AI gets the same statement confirmed from various sources, its trust in the detail rises markedly. Contradictions, by contrast, harm — make sure company name, location and services are written the same everywhere.
Ask ideally right after handover, when the joy over the finished furniture is fresh. A short hint like "If you're satisfied, a few sentences on Google would help us a lot, gladly with what we built for you" brings far better texts than an anonymous star rating weeks later.
A dedicated reference page per project
The strongest lever is a dedicated subpage for every larger project, instead of squeezing everything into an overcrowded gallery. A page titled "Solid-wood living kitchen in Fürth with seamless anthracite fronts" the AI can unambiguously assign to a topic, a location and a service. This focus is exactly what generative systems look for when they answer a location-related question.
On this page you bring in all five building blocks, supplemented by two to three captioned photos and, if available, the customer voice for the project. This way a closed proof arises: task, execution, material, result and confirmation in one place. Ten such pages are more valuable for your AI visibility than a hundred uncaptioned images in a single gallery.
Think here in your customers' questions. Someone doesn't type "furniture making", but "joiner for walk-in wardrobe roof slope" or "shop fitting bakery from oak". When one of your project pages describes exactly this situation, you become the fitting answer. Build your page titles and first sentences therefore from the customers' language, not from internal technical terms.
Common mistakes that devalue your references
The most common mistake is generalization. "We make individual furniture in premium quality" stands on every second joinery page and is worthless to the AI, because it contains nothing verifiable. Replace every such platitude with a concrete project with numbers. A single proven sentence beats ten advertising promises.
The second mistake is contradictions and outdated information. When the homepage says "since 1998", the imprint a different founding year, and Google still the old address, that unsettles the AI and it downgrades your information. Keep name, address, phone number and service description identical on all platforms. This consistency is unspectacular, but decisive.
The third mistake is giving no date at all. Projects without a year appear arbitrarily old. Write for every reference project when it was created. A "realized 2025" signals to the AI and the customer that your business is active and currently performs exactly this kind of work.
How to start this week
You don't have to do everything at once. Take your three best completed projects of the last two years and write down the five building blocks for each. That takes about ten minutes per project and immediately delivers you three real, provable reference texts that say far more than your entire previous gallery.
Then establish a routine: at every handover you dictate the five building blocks directly on site into your phone and ask the customer for a short, project-related review. This way your reference collection grows by itself, without you having to laboriously piece everything together at year's end. Out of every order automatically comes a building block of your AI visibility.
The effect doesn't come overnight, but it's sustainable. While ads only pay as long as you pay, well-documented references work for you permanently. Every described project stays a proof the AI can call up again and again when the next customer asks for exactly your kind of joinery work.
Common questions
Isn't it enough to simply upload lots of photos of my work?
No. Photos impress people, but AI systems like ChatGPT or Perplexity draw their answers almost exclusively from text. A picture of your oak kitchen stays invisible to the machine, as long as it doesn't say in words next to it that it's an oak kitchen, with material, dimensions and location. Add a concrete caption and a real alt text to every photo, then mute pictures turn into usable evidence.
Do I have to state prices for my reference projects?
No, that's not necessary and with individual joinery work often not sensible either, because every piece of furniture is made to measure. More important than the price are material, dimensions, task, special feature and timeframe. If you want to give a rough orientation, use a range like "fitted wardrobes from about 2,500 euros". For AI visibility, above all the concrete, verifiable details of your execution count, not the invoice total.
How many reference projects should I document?
Quality beats quantity. Ten cleanly described projects, each with its own page, clear material details and a customer voice, have a stronger effect than fifty short gallery entries. Start with your three best works of the last two years and then build a new reference at every handover. Pay attention to variety: kitchen, fitted wardrobe, staircase, shop fitting. This way you cover different customer questions and become the fitting answer for more search situations.
Read on