Ask the AI what it says about you. The answer is your first impression.

A restaurant owner from Cologne asks his guests how they found him. The answer used to be: googled, compared, called. Today one of them says: "I asked ChatGPT for the best Italian place in my neighborhood, and you came up."

I typed it in myself that evening. Same question. And his place wasn't there. The place two streets over was, the one that cooks half as well but has three times as many reviews and a head chef who gives interviews under his real name.

The guest got lucky. For him, the machine guessed right. For the next one, it guesses wrong.

That's the shift hardly anyone takes seriously.

Ten links were a chance. One answer is a verdict.

Google showed you ten blue links. Ten doors. You walked through them yourself, compared, decided. Whoever sat in position seven still had a chance if the headline landed.

An AI doesn't show ten doors. It names one to three names. It doesn't mention the rest. And the user never learns there was an alternative. That's the point that decides it: it's not that you rank further down. You're not there at all. For the guest who asks, you don't exist in that moment.

For a long time I thought this only hits the big categories. Wrong. It hits every question specific enough for a recommendation to make sense. "Best tax advisor for restaurant owners in NRW." "Renovation company that works fast and clean." "DJ for a wedding party with house music." For each of these questions, the machine builds an answer. And it builds it from everything that was ever online about you.

Why more content is exactly the wrong move

The obvious reaction once you get this: produce more. More blog posts, more Reels, more landing pages. Visibility through volume.

That's the reflex, and it makes things worse.

The AI boils everything it finds about you down to a few sentences. Your website, your LinkedIn, an old press article, a Reddit thread, a review from 2021. From all this material it forms a verdict. If your signals contradict each other, if the website says "Premium" and the reviews say "okay for the price" and the boss's LinkedIn profile has been dead for four years, the verdict gets blurry. And more content doesn't make a contradictory impression clearer. It makes it louder.

The old currency was the backlink. The analysis I looked at examined 82 factors that decide whether an AI recommends a brand. Backlinks landed at 1.9 out of 5 points. Almost meaningless. At the top were relevance and, nearly level with it, how often a name gets mentioned at all. Then reviews and authority. Third-party citations, meaning others write about you without you asking them to, scored high across six different platforms.

Translated, that means: it no longer matters how many pages link to you. What matters is how many people and sources tell the same story about you without you paying.

The mechanics: reputation is the ground everything stands on

You used to work on each channel separately. Website here, ads there, PR somewhere else again, and someone handles reviews on the side. Four construction sites, four people in charge, four truths.

That no longer works, because the machine reads all four at once and builds one single statement about you from them. And not only the search engine. The ad platforms now read your whole brand as input as well. Google's ad products scan website, landing page, product info and creatives to decide on their own what you offer and whom to target. If your channels say different things, that turns into noise, and the platform matches you with the wrong audience.

So reputation is no longer a PR discipline you tack on at the end. It's the infrastructure that search and advertising run on.

A side effect that makes this urgent: anyone who lands on your site through an AI recommendation arrives pre-vetted. The machine already did the comparing for them and convinced them. In the measurement cited, those visitors converted about eight times better than classic traffic. The number comes from an internal measurement by a single agency, more on that below, but the direction is plausible: people who are pre-sold buy more easily.

Implementation: the check you run today

You don't have to rebuild everything right away. First you need to know what the machine says about you. That's a ten-minute job.

1. Open the three big answer engines your customers turn to, one after the other.
2. Ask each one the same question, the way a customer would ask it. Not your company name. The category plus the region plus the deciding criterion.
3. Then ask directly about your name and have the AI describe your company in four sentences.
4. Copy the four sentences into a document, separately for each engine.
5. Mark each sentence with one of three colors: correct, outdated, wrong.
6. Count the red and yellow sentences. That's your first impression with every guest who asks.

For steps 2 and 3, use this prompt word for word:

A potential customer is looking for [category] in [region]
and cares most about [deciding criterion].
Which two to three providers would you name, and why?

Then: Describe the company [name] in exactly four
sentences, the way you would explain it to a customer
before first contact. State what you base your
assessment on.

The last sentence is the one that matters. "State what you base your assessment on" forces the machine to show its sources. That's where you see whether it reads your own website, old press, or a forum post you've never heard of.

Worked through for the Cologne restaurant: the owner asks, the result is four sentences. Sentence one, the cooking checks out, green. Sentence two, "solid value for money", yellow, because he positions himself as premium. Sentence three cites a review from three years ago about slow service, red, the problem was fixed long ago. Sentence four, "no well-known head chef", red, even though the man has worked in the same kitchen for twelve years. Three of the four sentences do damage. Not one of them is a lie. The machine only pieced together the wrong, old and missing signals.

The repair followed the building blocks that also came out on top in the analysis: the head chef goes public with his name, a local magazine profiles him, new reviews get an active nudge, and the website finally says clearly what premium means at his place. No new blog series. Fewer signals, but the right ones, and all of them tell the same story. Expert authorship, meaning a real name with proven expertise, scored over 4 out of 5 on all tested platforms in the measurement. Without public proof, the expertise just doesn't exist for the machine.

Where this gets overrated

I believe in the shift. I don't believe every number it gets sold with.

The eightfold conversion and the 82-factor analysis come from internal, unpublished measurements by an agency that offers a paid audit in the same breath. No description of methods, no sample size, no independent review. The bet is also the business model of the one who calls it. Take the numbers as a direction, not as a measurement.

Second: backlinks aren't dead. Classic Google search still sends traffic, and links still count there. If you tear down your backlink profile in a panic now, you cut into a channel that works. The right reading is this. For the AI answer, mentions count more. For classic search, both count. You shift your weight, you don't throw anything away.

Third, and this is the question nobody has a good answer to: how does a small business with no press history and no budget for its own studies realistically build a coherent reputation? For a niche with little content about you, the honest answer is: slowly, through real reviews and real names, not through original research you can't afford. The advice "produce less, but stronger" only helps if there's something there to strengthen in the first place.

The one sentence

Ask the machine today what it says about you, because that paragraph has long been your first impression, whether you wrote it or not.

FAQ

What is an answer engine, anyway?

An answer engine is an AI system like ChatGPT that answers a question directly with a finished answer instead of a list of links. It searches the sources available on the web, condenses them, and usually gives you only one to three names. Google search shows you ten results to compare yourself, while an answer engine makes the selection for you.

How do I find out what an AI says about my company?

You ask the AI the same question a customer would ask, meaning category plus region plus the most important criterion, without naming your company. Then you ask specifically about your company and have it described in four sentences. Also ask the AI to say what it bases its assessment on. That shows you the sources.

Don't backlinks count for SEO at all anymore?

They do. Backlinks aren't meaningless, they just count far less for AI recommendations than they used to. In the analysis cited in the article, backlinks reached only 1.9 out of 5 points on the question of whether an AI recommends a brand. For classic Google search they still matter, though. Only the weight shifts here, nothing becomes completely worthless.

What does expert authorship mean in this context?

Expert authorship means that a real person with a name and proven expertise is publicly and visibly tied to a company, for example through interviews or trade articles. An AI can only assess expertise if it finds public proof of it. Without a visible name backed by evidence, that expertise practically doesn't exist for the machine.

Is producing more content enough to do better in AI answers?

No, more content is often the wrong reflex here. An AI condenses all the information available about you into one verdict, and if that information is contradictory, the verdict doesn't get clearer, only louder. What matters more than volume is that website, reviews and press mentions tell the same consistent story.

Try it yourself

• npdigital.com, request an audit that shows how each AI engine currently describes your company. Keep in mind that this provider is the same one that collected the numbers cited here.