# A product profile built from your own website

> Demander reads up to eight pages of your public site and turns them into a product profile with pricing, proof, integrations and named competitors. Claims it finds wait for your approval before any copy can use them.

Paste the URL. Demander reads the homepage and the pages that say most about the product (pricing, features, customers, integrations, security, comparisons), then writes down what it found and on which page. Everything later in the plan is built on that profile, so you see it first.

## What happens after you paste the URL

1. **robots.txt first.** The crawler identifies itself as DemanderBot, reads robots.txt and stays out of anything it disallows. If the homepage itself is disallowed, the analysis stops and tells you why.
2. **The pages that describe the product.** From the homepage links it ranks internal pages: pricing first, then product and feature pages, solutions and use cases, customers and case studies, comparison pages, integrations, security and compliance, about and contact. It reads up to eight pages and skips the blog, careers, legal, login and cart pages, files, and pages with almost no text.
3. **A profile from the text.** A deterministic engine extracts the profile from the text, headings, buttons and JSON-LD of those pages. When AI is available, Claude reads the same pages and adds to the profile. The merge keeps every piece of evidence the deterministic pass found, so the model can add detail and cannot quietly drop any.
4. **Claims set aside for review.** Anything that reads as a claim (a number, a rating, a certification, a customer quote, a superlative) becomes a proposed claim with the page it came from. No copy can state it until someone approves it.
5. **Your design, read in the same pass.** The homepage stylesheets, fonts, logo and images are read on the way and become the website source of your [brand design](https://demander.to/features/brand-design/).

## What the profile contains

| Field | What Demander looks for | What it feeds |
| --- | --- | --- |
| Category and type | One of 14 software categories (or what a shop sells), the product type (SaaS, service, physical product, marketplace, app) and whether it sells to businesses or consumers | Which personas, offers and channels make sense |
| Pricing | Prices in EUR, USD, GBP, CZK and PLN with their period, plan names, the starting price and the model: free, freemium, trial, subscription, usage or quote only | Company size, contract value and the economics of each persona |
| Value and features | Value propositions, features, use cases and differentiators from headings and page sections | Pains, jobs to be done and the messaging pillars |
| Proof | Metrics, testimonials, ratings, certifications, logos and case studies, each with its source page | Proposed claims, proof messages and evidence scores |
| Integrations | Tools your site names, matched against about sixty common ones; the AI pass can add others the pages mention | Messages, objection answers and blog articles |
| Competitors | Names your pages put after phrases such as “vs”, “alternative to”, “compared to” or “migrate from”, in English, Czech and German | Search keywords and the competitor rules for copy |
| Buying complexity | Quote-only pricing, enterprise, SSO and procurement language and high prices, against a free plan, self-serve sign-up and “no credit card” | How the budget splits across funnel stages, which offers appear, how many personas |
| Targets and languages | Industries, roles, company sizes and countries the site mentions, and the language of each page | Default markets and persona firmographics |

Competitors come only from your own pages: the AI task is told to list only companies the site names as alternatives or comparisons, and integration partners are kept out of that list.

## Check it before anything is planned on top

When you add the URL, switch off **Plan everything after the analysis** and the run stops at the profile. With it on, personas, messages, journeys and first creatives follow straight away, all waiting for review; nothing launches without approval.

Corrections happen in three places. Proposed claims are approved or rejected on the Governance page, where you can also limit a claim to certain markets and channels, give it an expiry date, or add a claim your site does not show along with its evidence. When you generate the plan you choose the target countries and how many personas to propose, from one to four. And the profile text itself comes from your pages: change the page, run the analysis again, and the profile follows. A new run replaces the proposed claims and keeps the ones you already reviewed.

## How the profile keeps copy honest

- **Approved claims only.** Every metric, superlative, comparative or certification in generated copy must match an approved claim that covers the market and channel and has not expired. A statement with no matching claim blocks the creative, with a note on what to approve or remove.
- **Competitor policy.** By default your brand never names a competitor, and a creative that does is blocked. Switch the policy to allow comparative copy if your market expects it. Partners your site also lists as competitors are kept out of copy either way.
- **Every source listed.** The profile lists the pages it was built from, and each proof point and claim keeps the URL it came from, so a reviewer can check the original wording.
- **Works without AI.** The deterministic engine needs no AI and runs offline. It is also what answers when the month’s AI budget is used up, so planning never stops on a quota.

> **Sites that render in the browser.** The crawler reads the HTML your server sends. A site that fills in its text only with JavaScript gives it little to read, and the profile comes back thin with a low confidence score (the deterministic engine caps its own confidence at 0.85 even on a rich site). A pricing page and product pages with server-rendered text make the biggest difference.

## Where the profile goes next

The profile is the input to the [personas and buying committee](https://demander.to/features/personas-and-buying-committees/): pains come from your value propositions, company size from your pricing, markets from the countries and languages on the site. The [messaging](https://demander.to/features/messaging-and-funnel/) is written from the same facts, and the offer ladder depends on the buying complexity it found.

The free [snapshot](https://demander.to/snapshot/) runs this analysis on any public URL and shows the profile it found, with two personas, two messages, the funnel and three creatives.

## Questions

**Which pages does Demander read?**

The homepage and up to seven more internal pages, ranked by how much they say about the product: pricing, product and features, use cases, customers, comparisons, integrations, security, about and contact. It skips blogs, careers, legal and login pages, and it respects robots.txt.

**Can I fix something the profile got wrong?**

Claims are yours to approve, reject, scope or add. The descriptive fields (features, pricing, value propositions) are read from your site, so the fix is to change the page and run the analysis again. Countries and the number of personas are set when you generate the plan.

**Does it read my competitors’ websites?**

No. Competitors in the profile are the names your own pages use in comparisons and alternatives. They feed search keywords and the competitor rule for copy, which by default blocks naming them.

**Do I need my own AI key?**

No. Plans include an AI budget, and during the 14-day trial AI use is capped at $10. When AI is unavailable or the budget is used up, the deterministic engine builds the profile instead.

## Related

- [Personas and buying committees](https://demander.to/features/personas-and-buying-committees/)
- [Messaging and funnel](https://demander.to/features/messaging-and-funnel/)
- [Brand design](https://demander.to/features/brand-design/)
- [Free snapshot](https://demander.to/snapshot/)
- [How it works](https://demander.to/how-it-works/)

---
Source: https://demander.to/features/product-intelligence/ · Demander · reviewed 2026-10-02