The 10-Page SEO Starter Kit for a New AI Product
The ten pages a new AI product should have before writing anything else, what each one is for, and the order to build them in.
Published 2026-10-22 · 4 min read
Ten pages, built in order, is a complete SEO foundation for a new AI product. Not ten blog posts — ten functional pages, most of which you need anyway and which happen to be the ones that rank and get cited.
Building anything beyond these before they exist is premature. The list is short because the failure mode at this stage is volume, not scarcity.
The ten pages
| # | Page | Its job |
|---|---|---|
| 1 | Home / product | Defines the product in the first sentence and shows it working |
| 2 | Pricing | Answers the second question every visitor has; ranks for "[product] pricing" |
| 3 | One specific-job page | The narrowest description of the main use case; your first rankable query |
| 4 | One comparison page | You versus the alternative people actually evaluate you against |
| 5 | One definition page | What your category or approach is, defined in sentence one |
| 6 | Docs or how-it-works | The technical detail that answers real questions and earns trust |
| 7 | Changelog | Dated evidence of maintenance; each entry is a fresh indexable page |
| 8 | FAQ | Three to six real questions, answered in the first sentence of each |
| 9 | About | Who is behind it. Entity resolution and trust, both |
| 10 | Contact / support | A real route to a human. Reviewers and buyers both look for it |
Build order, and why
- 1
First: pages 1, 2, 9, 10
These make the site legitimate. Without them nothing else converts and some directories will not approve a listing. A day of work.
- 2
Second: page 3, the specific-job page
Your first genuinely rankable page, because the query is narrow enough that domain authority matters less. Write it from the words users actually used.
- 3
Third: page 4, the comparison
The highest-converting page type you can build, because the reader has already decided to buy something. Be honest about where the alternative is better.
- 4
Fourth: pages 7 and 8, changelog and FAQ
Cheap, and both produce indexable content continuously. The FAQ is also the most-quoted format in AI answers.
- 5
Fifth: pages 5 and 6
The definition page has the longest payoff and the highest ceiling. Docs accumulate value permanently.
The technical layer, once, across all ten
Do this before publishing rather than retrofitting. Each item is small and each one silently breaks something if missing:
- Canonical URL on every page, including the home page. Without it, parameter variants and trailing-slash duplicates compete with each other.
- metadataBase set, so relative Open Graph URLs are not silently dropped. In some frameworks this omission means every shared link renders with no image.
- Open Graph and Twitter card defaults in the layout, with per-page overrides for title and description.
- robots.txt permitting crawlers, including the AI retrieval bots, and disallowing only machine paths like API and checkout.
- A sitemap, submitted to search consoles. Chunked if the site grows.
- Server-rendered content. Anything that appears only after JavaScript runs may never be read.
- JSON-LD: Organization once sitewide, SoftwareApplication on the product page, Article on any long-form page, FAQPage where the questions are visible.
- sameAs on the Organization markup, listing every live profile. This is the entity-resolution signal and the most commonly omitted one.
What each page needs to do well
| Page | The one thing that decides whether it works |
|---|---|
| Home | Whether the first sentence defines the product. Not a value proposition — a definition |
| Pricing | Whether the numbers are real and the model is explained. Vague pricing pages lose the visit |
| Specific-job | Whether it matches the query the person typed. Title and H1 should contain their phrasing |
| Comparison | Whether it concedes something. A comparison with no concession is an advertisement and reads as one |
| Definition | Whether the first sentence stands alone. It is what gets quoted |
| Docs | Whether the answer is at the top. Nobody reads to the middle |
| Changelog | Whether entries say what changed for the user. "Bug fixes" tells them nothing |
| FAQ | Whether the visible text matches the markup. Divergence is a guideline violation |
| About | Whether a person is named. "A team of passionate builders" resolves no entity and earns no trust |
| Contact | Whether it works. Test it |
What not to build yet
- A blog with a publishing schedule. Ten functional pages first. A blog post is only worth writing when it answers a question you have actually been asked.
- Category head-term pages. "AI video generator" is held by domains with years of authority. You will not win it this year.
- Programmatic page generation. Thousands of templated pages is the pattern search policies penalise, and a domain-level penalty is far worse than having no traffic.
- Multiple locales. Until the first ten pages perform, translation multiplies maintenance without multiplying results.
- A second domain or a microsite. Split authority is worse than concentrated authority at this scale.
Measuring the ten
At ninety days, check three things per page rather than traffic totals:
- Impressions in search console. Is the page being considered at all? Zero impressions after ninety days means it is not indexed or not relevant.
- Which queries it appears for. Almost never the ones you expected, and the real queries are your next pages.
- Whether it converts. A page with modest traffic and good conversion deserves more internal links; one with traffic and no conversion needs rewriting.
Ten pages done properly is a complete foundation. Ten thousand done badly is a penalty risk, and the second option is the one most content advice quietly recommends.
More in SEO & GEO
- Entity Consistency: Why AI Models Fail to Attribute Your ProductWhen a model cannot tell your mentions describe one product, it attributes none of them. The signals that resolve an entity, and how to audit yours.
- How to Measure AI Referral Traffic When UTMs Are StrippedAI assistants strip UTM parameters, so referrer segmentation is the only reliable method. The setup, the domains to watch, and how to read the numbers.
- Backlinks for AI Startups: What Works When You Have No AuthorityEight link tactics ranked by effort and payoff for a domain with no authority, and the three that look productive and are not.