SEO for an AI startup: the job outlasts the label

Start with the pages that match how buyers search and how assistants cite. Keep it small. Ship a set you can improve.

• One job page per core task: “generate X from Y”, “detect X in Y”. Example: /use-cases/generate-sql-from-natural-language.
• A plain “What it does and what it does not” page. Limits, inputs, outputs, cost to run. Example: /what-we-do.
• A comparisons hub, with one page per rival and per incumbent. Example: /compare/yourproduct-vs-incumbent.
• An alternatives page that lists you and others a buyer weighs. Example: /alternatives/yourproduct-alternatives.
• A pricing page that states the model and the unit. Example: /pricing.
• An integrations page and a dated changelog. Example: /integrations, /changelog.
• If you sell to developers, a docs site with one task per page. Example: docs.yourproduct.com/getting-started.

These are the pages assistants quote. They answer the task, name the product, and give details a model can lift into an answer.

Your label will drift: “AI agent”, “copilot”, “autonomous workflow”, then the next term. Your buyer still needs the job done. Keep job pages stable.

1. 1

   Name the job, not the hype

   Use verbs and objects: “Summarise contracts to key clauses”. Put that in the H1 and the title. Keep the URL stable, like /use-cases/contract-summarisation.

2. 2

   Map synonyms to the same page

   Add a short section that names the current labels you hear. Example: “Also called AI copilot for legal review”. Do not split one job across five thin pages.

3. 3

   Refresh copy each quarter

   Update the opening lines and screenshots. Keep the canonical URL. Add internal links from blog posts on new terms to the stable job page.

4. 4

   Show inputs, outputs and limits

   Buyers and assistants quote details. Example: “Takes PDF up to 20 MB, outputs JSON schema ABC, average run time 18 seconds at 2 CPU cores.” Leave out what you cannot stand behind as of 2026.

A fixed page looks like /use-cases/email-personalisation with a clear H1, a short spec box, one walkthrough, and links to /pricing and /docs/api/send-email.

Generative engine optimisation is your main channel. Your buyers use assistants. You need to be named and linked in their answers.

• Allow the answering crawlers. Do not block Googlebot, Bingbot, OAI-SearchBot, Claude-SearchBot or PerplexityBot in robots.txt.
• Use consistent names. Product, plans, model names and team. Assistants match strings.
• Write lists and comparisons that can be quoted. Bullets, short tables, prices and limits in plain text HTML. Not hidden in images.
• Be on the lists assistants already cite. AI tool directories and roundups that rank for “best AI tool for X”.
• Measure by asking the questions a buyer asks. Read who is named. Track it monthly.

Add an llms.txt if you want to state terms for training. It does not replace robots.txt. It does not push you into answers.

Directories of AI tools rank on “best AI tool for X”. They are read by assistants. They refresh often. You need a presence you can maintain.

1. 1

   Pick the few that rank on your jobs

   Search “best AI tool for {your job}”. Make a list of the top 10 URLs you see twice a month. Add them to your outreach list.

2. 2

   Submit a tight listing

   Use your job wording. Include inputs, outputs, limits and pricing units. Link to /use-cases/job, /pricing and /what-we-do. Prefer your own domain over subdomains you cannot control.

3. 3

   Keep it fresh

   When your plan changes, update the directory page the same day. A stale price in a directory is what assistants repeat.

4. 4

   Earn a quote

   Offer one specific data point a curator will keep. Example: “Processes a mid-size CSV in 55 seconds on our mid plan”. Keep it honest as of 2026.

List pages on your site help too. Ship /lists/best-ai-tools-for-{job}, but make it useful. Add criteria, short reviews, and disclose you built one of them. Do not copy directories or you risk thin content that never earns links.

Your buyer searches by comparison. They also ask what your product can and cannot do. These pages convert and are cited by assistants and reviewers.

• Ship /compare/yourproduct-vs-incumbent with a clear table: use case fit, inputs, outputs, hosting, data retention, price unit.
• Ship /alternatives/yourproduct-alternatives. List real rivals. Say where each fits. Do not smear. Say when another tool is a better fit.
• Ship /what-we-do. A plain page with sections: problems we solve, who it is for, who it is not for, limits, privacy, pricing model.
• Use the same names across all pages. Assistants rely on exact strings when they pull facts.

A fixed comparison page looks like /compare/yourproduct-vs-bigtool with a three row table and a short verdict for three common buyer roles.

Developers search by task, by error, by comparison and by alternative. Documentation ranks. Assistants cite docs. Treat docs as product pages.

• One page per task. H1 is the task. Answer first, then code. Example: docs.yourproduct.com/ingest-json.
• A getting started that runs in five minutes. README snippets mirrored on the docs page, not only on GitHub.
• Package registry pages for your SDKs. Link them back to docs. npm, PyPI, crates.io and Docker Hub rank on the SDK name.
• An integrations page and a dated changelog. These earn long tail searches.
• Error guides that match your messages. Example: docs.yourproduct.com/errors/E1234-timeout. Assistants surface these when a developer asks by code.

Put docs in plain HTML. Do not hide content behind tabs that need JavaScript to render. AI crawlers read HTML. Assistants quote what they can fetch and parse fast.

You are small. You need a handful of clean links and brand mentions. Earn them where early adopters read and where assistants crawl.

• Launch on Product Hunt, Hacker News, Indie Hackers, BetaList and the small launch sites. These give you the first links and press mentions.
• Post a clear build thread on X and in the few communities where your buyer lives. Link each update to a page that should rank.
• Pitch two case studies, even if small. Name the customer, the job and the outcome. Assistants lift named companies.
• List on review sites if you are B2B SaaS. G2 and Capterra rank for “best X software” searches.
• Pitch one newsletter that covers your job. Ask for a link to the job page, not just the homepage.

Do not spray guest posts. You do not need volume. You need ten clean, relevant links that point to the pages above. Track them in your own sheet. Update anchors if the wording changes.

Finish the craft on each page. Assistants and people both need clarity. Keep it tidy and fast.

• Title around 60 characters. Lead with the job or the comparison. Example: “Generate SQL from text for analysts”.
• Meta description up to 155 characters. Write a plain sentence with the job and the next step.
• H1 matches the job or the rival name. One H1 per page.
• A spec box near the top with inputs, outputs, limits and hosting. Plain HTML, no image text.
• Internal links from the homepage, /use-cases and /blog to these pages. Use exact match anchors when it reads well.
• Add SoftwareApplication or Product schema where it fits. Keep it truthful. Use JSON-LD in the HTML.
• Images are compressed and have real alt text. Use lazy loading below the fold.
• Page speed: aim for LCP at or under 2.5 seconds, CLS under 0.1, INP under 200 ms.

A fixed page looks like /what-we-do with a 12 line summary, a four point spec box, one pricing link, one docs link, and a short FAQ at the end.

Robots controls are voluntary, but they are the only switch you have. If you block the answering and search crawlers, you will not be cited. Keep them open unless policy forces you to block.

Training crawlers have separate tokens. Blocking them does not remove pages already collected. It does not affect Search ranking. Set your policy, then move on.

You win by shipping and refreshing. Set a light cadence. Measure three things and act each month. Every quarter, re-run the plan.

1. 1

   Track buyer queries in assistants

   Ask ChatGPT, Perplexity, Google AI Overviews, Claude and Copilot the real questions a buyer asks. Log who they cite. Aim to be named on three job queries and one comparison.

2. 2

   Read Search Console for the new terms

   Watch the Performance report for rising impressions on new labels. Map them to a stable job page. If intent shifts, adjust the copy, not the URL.

3. 3

   Refresh the pages that earn demand

   Update /what-we-do, the spec boxes and the comparisons. If you ship a new integration or plan, reflect it the same day. Add one new job page only when it is a real job.

4. 4

   Maintain the lists and directories

   Update your top five directory listings. Keep your own /lists pages fresh. Remove dead tools. Add new names your buyers mention.

If a page gets impressions but never clicks, fix the title and first lines. If a job page splits traffic with a blog post, link the post to the job page and make the intent clear. If two pages target the same job, merge them and redirect.

Here are five concrete pages that reflect this plan. Use your own words and details. Keep the URLs stable.

• /use-cases/generate-sql-from-natural-language. H1: Generate SQL from text. Spec box: inputs, outputs, limit, average run time, data retention. CTA to /docs/sql and /pricing.
• /what-we-do. Sections: Problems we solve, Who it is for, Who it is not for, Limits, Privacy, Pricing model. One table that maps plan names to caps.
• /compare/yourproduct-vs-excel. Table rows: Data size, Automation, Auditability, Team features, Price unit. Verdict by role: analyst, engineer, manager.
• /alternatives/yourproduct-alternatives. Five tools, one paragraph each. When to pick them, when to pick you. A closing FAQ.
• /lists/best-ai-tools-for-contract-summarisation. Criteria, short reviews, disclosure. Links to each vendor and to your /use-cases page.