Preferly
Free check · private preview

When customers ask AI for a recommendation, does it name you?

We ask ChatGPT, Gemini and Claude the questions your customers ask. Then we show you whether they name you, who they recommend instead, and where they contradict your website.

Example · Riverside Family DentalFictional business

Done. 40 answers read, and the result for Riverside Family Dental is ready.

ModelQuestions 1 to 4Latest
ChatGPT (GPT-6.1 Sol)2 named you, 2 named othersNamed you
Gemini1 named you, 3 named othersNamed you
Claude Opus2 named you, 2 named othersNamed you
Grok2 named you, 2 named othersNamed you
Meta Muse Spark4 named othersNot you, 3 others
DeepSeek2 named you, 2 named othersNamed you
Qwen1 named you, 3 named othersNamed you
Kimi4 named othersNot you, 4 others
NVIDIA Nemotron2 named you, 2 named othersNamed you
GLM2 named you, 2 named othersNamed you
Answers read
40 of 40
Times named
14
Recommendations
169
Preferly score46 out of 100

An example of what you'll see while your check runs. The business and its results are made up.

Check your business

Helps us find the right business.
We’ll send a 6-digit code to start the check.

We'll email your report and occasional product updates. Unsubscribe anytime.

AI modelsin the full report, asked with web search on
10
AI agentseach with a label saying exactly how it's covered
5
Customer questionsin the free check, put to 3 models
5
About a minutefor the free check, while you watch
60 s
What you get

Start with the free check

The free check puts 5 customer questions to 3 AI models. The full report asks 10 models and tells you what to fix first. AI agents are covered separately, and we say exactly how.

The free check compared with the full report
What you getFree checkFreeFull report$99 one-time
Who we ask3 AI models10 AI models (AI agents: see how each is covered below)
Your scoreOut of 100, with its four partsThe same score, from all 10 models
Does AI name youHow often, across every answerPer model and per question, with rank
CompetitorsHow many are named instead of youWho they are, and the sources AI cited for them
Fact checkUp to 2 places AI contradicts your websiteEvery claim AI made about you, with the evidence
AI accessOpen or blockedPer crawler and agent, and whether an agent can reach your booking or checkout page
What to fix—A work plan in order, with the steps for each fix
DeliveryOn screen in about a minuteAbout 5 minutes, on screen and by email, with a PDF

Unlock the full report from your free result. AI agent results aren't part of it yet: Hark and Meta Muse are asked by hand, separately.

Who we ask

Models are asked through their APIs with web search on. Those answers approximate, but aren’t identical to, what each app shows its users. AI agents have no public API, so each one is covered differently, and we always say how.

Free check · 3 models
  • ChatGPT (GPT-6 Luna)
  • Gemini Flash
  • Claude Sonnet
  • ChatGPT (GPT-6.1 Sol)
  • Gemini
  • Claude Opus
  • Grok
  • Meta Muse Spark
  • DeepSeek
  • Qwen
  • Kimi
  • NVIDIA Nemotron
  • GLM
5 AI agents · how each is covered

Agents asked by hand (Hark and Meta Muse) aren’t part of the automated full report yet. We’ll say when they are.

  • HarkAsked directlyOur team asks your questions in the Hark app from a standard account with memory off, one fresh thread per question.
  • Meta MuseAsked directlyOur team asks your questions in the Muse app from a standard account with memory off, one fresh thread per question.
  • OpenAI dotsVia the GPT model behind itWe don't ask dots itself. We use the answers from the OpenAI GPT model that powers it.
  • xAI Grok BotVia the Grok model behind itWe don't ask Grok Bot itself. We use the answers from the Grok model that powers it.
  • InstinctNot tested — awaiting permissionInstinct's terms don't allow this kind of testing. We've asked for permission and won't test it until we have it.
What the report shows

A score, the reasons, and a plan to fix it

From a real check we ran, with the name left out. A developer-tools startup we checked: 11 AI answers to customer questions.

  1. Named in 3 of 11 AI answers

    When customers asked for a recommendation without naming the business, most answers named someone else.

  2. 1 competitor recommended instead

    Businesses that AI recommended more often and ranked higher. The full report names them.

  3. 1 place where AI contradicts its website

    The result puts what AI said beside the quote from the business’s own page.

Preferly score
56 out of 100
Visibility
10.9 / 40
Accuracy
23.7 / 25
Sources
2.7 / 15
Access
18.7 / 20
  1. Do firstWO-01Give AI a source about you it can citeup to 4 pts · Medium effort
  2. WO-02Publish pages that answer the 2 questions you're missing fromup to 3 pts · Medium effort
  3. WO-03Get included on the sites AI reads for your categoryup to 2 pts · Medium effort
  4. NextWO-04Check your services: AI and your website disagreeup to 1 pts · Low effort

From a check run on 19 Oct 2026. AI answers change from run to run, so a result is a dated snapshot.

How it works

Three steps, about a minute

You give us your business name. We do the rest while you watch.

  1. We find your business

    From the name, and your website if you give it, we confirm your site and work out what customers come to you for.

  2. We ask AI what your customers ask

    5 questions go to ChatGPT, Gemini and Claude, with web search on. Most don’t mention you by name, the way a new customer would ask.

  3. We read every answer

    We count who was recommended, check what AI said about you against your own website, and test whether AI tools can reach your site.

Method and trust

Questions people ask

The short answers. The full detail is on the methodology page.

Is this the same as asking the ChatGPT app?

Close, but not identical. We ask each model through its API with web search on and a neutral instruction. The apps add their own instructions, personalization and memory, so a signed-in user may see a somewhat different answer.

Each question is asked once per model, and answers vary from run to run, so treat a result as a snapshot. Every result shows its date and the exact questions.

How is the score worked out?

Four parts, 100 points in all.

  • Visibility (40): how often AI recommends you when the question doesn’t name you, weighted by rank.
  • Accuracy (25): the share of checkable things AI said about you that match your website.
  • Sources (15): how often AI cites your own site or profiles.
  • Access (20): whether AI tools are allowed in, can load your site, and can read your key facts.
Why would AI contradict my website?

AI answers mix what a model learned in training with what its web search found: old directory listings, review sites, articles, sometimes an old version of your own site. We only flag a contradiction when we can quote the part of your site that disagrees, so you can see which source needs updating.

Do you store my data?

We keep a record of each check, with its answers and result, and the email address you gave us, so we can run the service, fix problems and send you your report. We don’t sell it or send marketing without asking. To run the check we send your business name and website to the AI models listed above, and we read public pages of your site.

Do you pretend to be another company’s bot?

No. Our crawler says it is Preferly, never gets around bot protection, and never submits a form. If your site blocks automated visitors, we report that instead of working around it.

Preferly

Know when AI recommends you, and when it doesn’t.

Run the free checkFree, and it takes about a minute.