AI Visibility

Open AI Visibility

AI Visibility, the PirateSERP tool for AI search presence, asks ChatGPT, Claude, Gemini, and Perplexity the same prompt and shows each answer, the pages each model cites, and whether your brand appears. The models search the web as they answer, so each answer carries its sources. Asking the models does not require your own AI key. To see how each answer describes your brand, add your OpenRouter API key and model in Account Settings.

Ask a prompt

1. In the header, select the project. Your business name fills Brand to look for (optional), and your project's market fills Web search country when that country is on the list.
2. In the Prompt box, enter a prompt of up to 500 characters.
3. For Models, select the models to compare: ChatGPT, Claude, Gemini, and Perplexity. The form selects all four by default.
4. Optional: For a mention check, in the Brand to look for (optional) box, check the brand name.
5. In the Web search country list, select a country, or keep Default. The list covers 24 countries, including the United States, the United Kingdom, Canada, and Australia. Gemini answers without a country setting.
6. Click Ask the models.

Answers take up to two minutes, plus a short framing check when you have an OpenRouter key. Asking the same prompt again within a week returns the saved answers.

Write a prompt that your customers could ask, such as Which accounting tools support a small construction company?

Read the answer cards

Each model has its own card. A card shows the model's answer, its Sources, the Searches the model ran, and the model name.

• Sources lists the pages the model cited, with each page's title and domain. Open a cited page to check what supports the answer.
• Searches the model ran lists each web search query the model sent while it wrote the answer, with repeated queries shown once.
• The mention badge, Mentions or No followed by your brand, checks the brand text against the answer and its cited pages. A cited page that matches your brand appears in bold.

An answer can contain no sources. When one model fails, its card shows an error, and the other cards keep their answers.

One prompt samples one response from each model. It does not measure every customer search or establish a fixed ranking in an AI assistant. Write a different prompt to test a different customer need, rather than treating a repeated answer as an independent sample.

Read the visibility summary

When you enter a brand, a summary above the answer cards shows how far your brand got in each answer. Being cited and being recommended are different results: a citation means the model found your content useful to consult, and a recommendation means the rest of the web backs your brand strongly enough for the model to endorse it.

The summary table has one row for each model that answered:

• Cited shows whether the model cited a page that matches your brand.
• Named in answer shows whether the answer text names your brand.
• Framing shows how the answer describes your brand: Recommended, Neutral, Hedged (named with caveats or doubts), or Recommended against. The framing check uses your own OpenRouter key and model, with one request per prompt. Without a key, the table has no Framing column, and a note explains how to turn the check on.

The sentence above the table names the biggest gap. For example, Cited by 3 of 4 models but recommended by 0 often means that your pages are useful sources, but other sites don't corroborate your brand yet. Earn mentions, reviews, and comparisons on other sites to close that gap. When a model recommends against your brand, read that answer and its sources first.

Framing is an estimate of one sampled answer. Read the sentence that names your brand before you act on it. Past prompts saved before the framing check was added show the table without the Framing column.

Reopen a prompt

Past prompts lists every prompt that you asked in the active project, with its date and time. Click an entry to review its saved answers, or click Delete run to remove it. Opening a past prompt while another prompt runs does not cancel that run; the running prompt still finishes and appears in Past prompts.

When you asked the same prompt earlier in the same project, the reopened prompt shows Changes since followed by the date of the earlier run. Letter case and extra spaces don't make two prompts different. For each model that answered both times, the list shows what changed:

• Whether the answer names your brand, for example PirateSERP named: No → Yes.
• How the answer frames your brand, when both runs looked for the same brand.
• New sources: sites that the model cites now but didn't cite before.
• Sources dropped: sites that the model cited before but doesn't cite now.

A model with nothing to report shows No change. Ask a prompt again after you publish or earn new coverage to see whether the answers moved.

See which sites the models cite

When the active project has saved prompts, the Sources the models cite section combines the sources from every saved answer in the project, up to your 100 most recent prompts. The table lists each site with these columns:

• Type: Review site, Forum, Wikipedia, Video, or Editorial / other.
• Citations: how many times the models cited a page on that site.
• Share: that site's part of all citations.
• Without followed by your brand: how many of those citations came in answers that don't name your brand.

The table shows the 10 most cited sites. To see the rest, click Show all.

Sites to earn a mention on lists the sites that the models cite most often in answers that leave your brand out. A review, listing, or discussion of your brand on those sites gives the models a source that names you. The section uses your project's business name as your brand. If the project has no business name, it uses the brand from your most recent prompt that had one. If neither exists, add a business name to the project to see the gap list.

See cited sites and named entities

The Cited sites and named entities section works from the same saved answers, up to your 100 most recent prompts. It uses no new searches, so it costs nothing to open.

• Cited sites lists each site the models used as a source. The Role column marks your own site, or a competitor whose name matches the site's name. That match is by name, so check it before you act on it. Click a site to see each page it cites, with its title and the model and time of each citation. A citation means the model used the page as a source. It does not mean the model endorses the site.
• Each prompt lists every saved prompt. For each model it shows whether the answer Mentioned your brand, Cited as source (a cited page matches your brand), both, or Not mentioned it. The sentence that names an entity is quoted from the answer. Competitor names in the answer are listed the same way. When a model is checked, a name that belongs to a different business or thing with the same name (a plumber called Apex and a game called Apex Legends, say) is listed as Same name, different entity and is not counted as a mention. The section's Name matches line says whether the check ran: Checked by model, or Name match only when it did not. The check needs your project's website and industry, an English project, and your own OpenRouter key.
• Change since previous run compares a prompt with the last time you asked it. The comparison runs only when the same models answered both times. If the models differ, the section says the runs are not comparable and lists no changes. Otherwise it lists each new or lost mention by model.

Compare your brand with competitors

The Share of voice section shows how often each model names your brand and your competitors across your saved answers.

1. In the Competitors (optional) box, enter up to five competitor names, separated by commas. For example, Semrush, Ahrefs, Moz.
2. Click Compare.

The table has one row for each brand and one column for each model. Each cell shows:

• The share of that model's answers that name the brand, with the count, for example 50% (2 of 4).
• The average Position of the brand among the brands in the table. Position 1 means that the answer named that brand before the others. Not named means that no answer from that model names the brand.

A name matches only as a whole word, and letter case doesn't matter, so Beta doesn't match Betamax. A name that the model's answer uses for a different business is not counted, and the Name matches line under the table says whether that check ran. Enter each competitor's name the way the answers write it. Your competitor list stays in this browser for the active project; it isn't saved to the project and doesn't change your project profile.

Continue your research

Check the pages that the models cite against your own site with Domain Overview and Backlinks. Define your brand as an entity with Entity Home, and add structured data with Schema Generator.