Sealyx documentation

Everything the product does, in the order you will meet it: run a free growth diagnosis from a URL or a GitHub repository, read the five growth scores, work the action queue, and connect Search Console and GA4 when you want real traffic in the loop.

Run a free audit See pricing

What Sealyx is

Sealyx is a growth operating system for people who ship software alone. It answers one question: what should I do today to get the next batch of users?

Most SEO tools hand you a list of 200 issues and leave the prioritisation to you. Sealyx inverts that. It reads your live site or your source code, scores five growth dimensions, names the single bottleneck holding you back, and turns it into an ordered queue of actions with copy-paste implementation steps.

Two surfaces matter today: classic search (Google, Bing) and AI answers (ChatGPT, Gemini, AI Overviews). Sealyx measures both, because buyers increasingly ask an assistant for a recommendation instead of scrolling a results page. Being invisible in AI answers is the new page-two.

Nothing in a report is a template. Every number comes from a request made while you waited: an HTTP fetch of your page, a fetch of robots.txt, sitemap.xml and llms.txt, a parse of the server-returned HTML, and live prompts sent to AI assistants. Where a value cannot be measured, the report says unknown instead of guessing.

What happens after you press Audit

01
Fetch
A single request to your URL with a normal browser user agent. Sealyx records the status code, every redirect hop, the final URL, response time, server header, compression and payload size.
02
Read the server HTML
The response body is parsed before any JavaScript runs — the same view a crawler or an AI agent gets. Title, meta description, canonical, H1 and heading tree, image alt text, internal links, structured data and word count are extracted from it.
03
Fetch the machine files
/robots.txt, /sitemap.xml and /llms.txt are requested separately. Robots rules are evaluated per crawler, including GPTBot, ClaudeBot, PerplexityBot and Google-Extended; the sitemap is parsed for URL count and last-modified dates.
04
Probe the AI surface
Sealyx derives buyer questions for your category from what the page says, sends them to live assistants, and stores the verbatim answers plus which brands were named.
05
Score and rank
Findings are folded into the five growth dimensions, the lowest-scoring dimension with real upside becomes the bottleneck, and actions are ordered by impact, confidence, speed and effort.
06
Attach a playbook
Every action is paired with an implementation page pre-filled with your own values, so the fix is copy-paste rather than research.

Quick start — 40 seconds to a diagnosis

There is no setup, no tracking script and no credit card for the first report.

1. Enter your URL: On the home page, paste your domain and press Audit my site. Sealyx crawls the page, robots.txt, sitemap.xml and llms.txt, then asks live AI assistants what they recommend for your category.
2. No domain yet?: Click No Domain under the input and connect a GitHub repository instead. Sealyx reads your source (public or private) and runs the same checks against the code that will ship.
3. Read the diagnosis: The report opens with the Next Best Move, then the five-dimension Growth Diagnosis, then the evidence: AI visibility, technical SEO, and code findings.
4. Work the queue: Each action carries impact, confidence, speed and effort. Open Steps & example code for a ready-to-paste implementation.

Two inputs, one pipeline

A live URL and a GitHub repository are two ways to produce the same thing: a set of site facts. Everything downstream is identical.

The URL adapter fetches the rendered HTML, follows redirects, and records the server-visible text. The GitHub adapter reads route files, metadata definitions, static assets and configuration to predict what a crawler would see after deploy. Both normalise into the same shape, so a project can start as code and later become a domain without losing history.

Private repositories are supported through GitHub authorisation. Access tokens are encrypted at rest and only used to read the files needed for the audit.

Growth Diagnosis — the five dimensions

Every scan produces five scores out of 100 and one bottleneck. The bottleneck decides what gets recommended first.

A dimension score is not a vibe. Each dimension owns a fixed set of checks; each check returns pass, partial or fail with the measured value attached, and contributes a weight to its dimension. A check that cannot be measured — for example Discovery without Search Console — is excluded from the denominator rather than counted as zero, so the score never punishes you for data you have not connected yet.

The bottleneck is the dimension with the lowest score among those with meaningful headroom, tie-broken toward the one whose failing checks are cheapest to fix. That is why a site with a beautiful landing page can still be told to ship a sitemap first: Conversion cannot pay off while Discovery is delivering nobody.

How a score is produced

01
Collect facts
Every fact from the crawl, the machine files, the AI probes and — when connected — Search Console and GA4 is written into one normalised record.
02
Run the checks
Each check reads only from that record, so the same input always yields the same verdict and you can see the exact value behind every result.
03
Weight and roll up
Checks are weighted by how much they historically move traffic; unmeasurable checks drop out of the denominator.
04
Pick the bottleneck
Lowest score with real headroom wins; the report then explains in one sentence why that dimension is limiting you.
05
Order the queue
Actions belonging to the bottleneck float to the top, then the rest sort by impact over effort.

Discovery: Can people and crawlers find you at all — indexability, sitemap, internal links, ranking surface, impressions from Search Console when connected.
Conversion: Once someone lands, does the page explain the product and lead to an action — headline clarity, above-the-fold copy, calls to action, GA4 engaged sessions when connected.
Authority: Signals that make an engine trust you — structured data, consistent naming, third-party mentions, backlink surface.
Distribution: Where you show up off-site — directories, launch platforms, communities, comparison pages.
AI Visibility: Whether assistants name you when a buyer asks for a recommendation in your category, and who they name instead.

GEO — generative engine optimisation

GEO is the practice of getting named inside AI answers. Sealyx runs it as a live measurement, not a guess.

Sealyx derives the buyer questions someone in your category would actually type, sends them to AI assistants, and records the verbatim answers. It then extracts whether your brand appears, in what position, and which competitors are recommended instead.

The fix side is concrete: AI crawlers must be allowed in robots.txt, key claims must exist as server-rendered text rather than client-only JavaScript, and your category, pricing and differentiators must be stated in machine-readable form.

Two things decide whether an assistant can name you. First, retrievability: its crawler has to be permitted and has to find real text, not an empty shell that fills in after hydration. Second, quotability: the answer engine needs a short, factual, attributable statement it can lift — what the product is, who it is for, what it costs. Marketing adjectives are not quotable; a sentence like "a $9/month uptime monitor for solo developers" is.

The GEO measurement, step by step

01
Infer your category
Sealyx reads your title, H1, above-the-fold copy and structured data to decide what you would be shortlisted for — for example "AI SEO tool for indie developers" rather than your brand name.
02
Generate buyer questions
Four to eight prompts a real buyer would type are produced from that category: best-of questions, alternative-to questions, and a use-case question. Brand names are deliberately left out so the answer is a genuine recommendation.
03
Ask live assistants
Each prompt is sent to ChatGPT and Gemini in a fresh session with no personalisation, and the full answer text is stored with the report.
04
Extract mentions
Answers are scanned for your brand and domain, for the position you appear in, and for every competitor named. That produces the mention rate and the competitor set.
05
Check retrievability
In parallel, robots.txt is evaluated for GPTBot, ClaudeBot, PerplexityBot and Google-Extended, and the server HTML is measured for how much of your key copy exists before JavaScript runs.
06
Check quotability
llms.txt, structured data, an explicit category sentence and visible pricing are each verified; whatever is missing becomes an action with generated content ready to paste.

Mention rate: How often you appear across the generated buyer questions.
Named instead: The competitor set that takes your slot, so you know who to write comparison pages against.
Crawler access: Whether GPTBot, ClaudeBot, PerplexityBot and Google-Extended are allowed.
llms.txt: A plain-text summary of your product for language models, checked and generated for you.

Technical SEO checks

Real requests, not a checklist. Each check reports the current value, why it matters and how to fix it.

Every SEO verdict below is derived from the raw HTTP response, never from a cached third-party index. That matters for a young site: an index-based tool tells you what Google saw weeks ago, while Sealyx tells you what is true right now, including on a preview deployment that has never been crawled.

Checks are reported with their measured value — the actual title and its length, the actual canonical target, the number of URLs in your sitemap — so you can verify the finding yourself instead of trusting a score.

The SEO pass, step by step

01
Resolve and fetch
Follow redirects from the URL you entered to the final one, recording each hop. Multiple hops, a http→https→www chain, or a 4xx/5xx ends the pass with a critical finding.
02
Evaluate crawl rules
robots.txt is parsed and matched against your URL for Googlebot, Bingbot and the AI crawlers; meta robots and X-Robots-Tag headers are read for noindex and nofollow.
03
Parse the document
Title, meta description, canonical, Open Graph and Twitter tags, the heading tree, image alt attributes, the html lang attribute and every internal link are extracted from the server HTML.
04
Measure render depth
The word count of the server response is compared with the text present after hydration. A large gap means crawlers and AI agents read far less of your page than your visitors do.
05
Validate machine files
sitemap.xml is fetched and parsed for URL count, last-modified freshness and dead entries; JSON-LD blocks are parsed and validated for required properties per schema type.
06
Report with values
Each check returns pass, partial or fail with the value found, the reason it matters, and a playbook containing the corrected markup for your page.

Indexability: Status code, redirects, robots.txt rules, noindex directives, canonical target.
Metadata: Title length and keyword placement, meta description, Open Graph and Twitter card completeness.
Structure: A single H1, heading hierarchy, image alt text, internal link depth.
Machine readability: sitemap.xml presence and freshness, JSON-LD structured data, language declaration.
Render depth: How much real text exists in the server response versus after hydration — the number AI crawlers actually see.
Performance surface: Response time, server, compression and payload signals collected during the crawl.

Playbooks — every finding has an implementation page

Recommendations without code are homework. Every action in Sealyx opens a playbook with the goal, ordered steps, an example you can copy, and a way to verify the fix.

Playbooks are pre-filled with your domain, your title and the values found during the scan, so the snippet is usually paste-ready. After shipping, re-check the site and the action moves to done with a before/after snapshot.

Example: generated llms.txt

# Your Product
> One-sentence description of what you do and who it is for.

## Product
- Category: <what buyers would search for>
- Pricing: <plans and starting price>
- Best for: <the specific user>

## Docs
- https://yoursite.com/docs
- https://yoursite.com/pricing

Monitor — snapshots over time

Monitor is the signed-in console. Add a domain and Sealyx stores a full snapshot on every check so you can see what changed.

Stored per snapshot: Resolved IP addresses, final URL, status code, response time, server, language, title, description, H1, canonical, social preview, favicon, word count.
Machine files: robots.txt, sitemap.xml, llms.txt, structured data, AI crawler rules.
History: Compare any two checks to confirm a fix landed or catch a regression after a deploy.

Connect Search Console and GA4

Static checks tell you whether the site is healthy. Real data tells you whether you are growing.

Search Console: Impressions, clicks, CTR and average position for the last 28 days, plus top queries and pages. Surfaces quick wins ranking in positions 8–20.
Google Analytics 4: Sessions, engaged sessions, conversions and top landing pages, so optimisation targets pages that actually convert.
Effect on scoring: Discovery and Conversion switch from heuristic estimates to measured values, and the Next Best Move starts optimising for return per hour.

The GEO playbook — how to get named in AI answers

Getting mentioned by an assistant is not luck. It is the result of four things being true at once: the crawler can reach you, the text exists without JavaScript, your category is stated plainly, and your facts are short enough to quote.

Start from the buyer's question, not your brand. Nobody asks an assistant "tell me about Sealyx"; they ask "what is a good SEO tool for a solo developer?". The answer to that question is assembled from pages that state, in plain sentences, who a product is for and what it does. If your homepage says "ship faster, grow smarter", there is nothing for the model to attach to that question.

Then make yourself retrievable. Allow GPTBot, ClaudeBot, PerplexityBot and Google-Extended in robots.txt, and confirm that your key claims are present in the server response — view source, do not inspect the DOM. A single-page app that renders everything client-side is invisible to most agents even though it looks perfect in a browser.

Finally, make yourself quotable. One sentence naming the category, one naming the user, one naming the price. Structured data for Product, Organization or SoftwareApplication. A short llms.txt at the root. Comparison and alternative pages, because those are the pages assistants pull from when a buyer asks for options.

A four-week GEO run

01
Week 1 — measure
Run a scan to record your current mention rate and the competitor set that appears instead of you. Without a baseline you cannot tell whether anything you do next works.
02
Week 1 — unblock
Fix robots.txt so AI crawlers are allowed, and move any key copy that only exists after hydration into the server response.
03
Week 2 — state the category
Rewrite the title, H1 and first paragraph so a stranger can tell what the product is, who it is for and what it costs in five seconds. Add Product or SoftwareApplication structured data with the same facts.
04
Week 3 — publish quotable pages
Ship an alternatives page, a use-case page and a pricing page with real numbers. These are the pages assistants cite when a buyer asks for a shortlist.
05
Week 3 — add llms.txt
Put a short machine-readable summary at /llms.txt with category, pricing, best-for and links to your docs.
06
Week 4 — re-measure
Re-run the same prompts and compare mention rate and position. Keep what moved, drop what did not.

robots.txt that keeps AI answers open

User-agent: *
Allow: /

User-agent: GPTBot
Allow: /

User-agent: ClaudeBot
Allow: /

User-agent: PerplexityBot
Allow: /

User-agent: Google-Extended
Allow: /

Sitemap: https://yoursite.com/sitemap.xml

The technical SEO playbook for small product sites

Most small sites do not lose traffic to competition. They lose it to five preventable mistakes, all of which are visible in the first HTTP response.

The five: content that only exists after JavaScript; a canonical tag pointing at the homepage on every page; a leftover noindex from staging; no sitemap, so new pages wait weeks to be discovered; and duplicate titles across routes, so search engines cannot tell your pages apart. Each takes minutes to fix and each can suppress an entire site.

Order matters. Indexability comes first — a beautifully written page that is blocked earns nothing. Then discovery: a sitemap and internal links so crawlers can reach every page. Then relevance: unique titles, descriptions and one H1 per page that match what people actually search. Only then is it worth investing in more content.

Internal links do more work than most people expect. A page linked from your navigation and footer is crawled far more often than an orphan reachable only from a sitemap, and the anchor text tells engines what the destination is about. Link with the words people search for, not with "click here".

Fix order that actually works

01
1. Make it indexable
Check robots.txt, meta robots and X-Robots-Tag for anything blocking the page, and confirm the URL returns 200 with at most one redirect hop.
02
2. Make it renderable
View source and confirm your headline, value proposition and main copy are present in the raw HTML. If they are not, server-render them.
03
3. Make it discoverable
Publish sitemap.xml with every public URL, reference it from robots.txt, and link every important page from the header or footer.
04
4. Make it distinguishable
Give each route a unique title under 60 characters, a description under 160, exactly one H1, and a self-referencing canonical.
05
5. Make it understandable
Add JSON-LD matching the page type, alt text on meaningful images, and descriptive anchor text on internal links.
06
6. Keep it that way
Re-check after every deploy. Regressions are the norm, not the exception, and they are silent.

Self-referencing canonical and unique metadata

<title>Pricing — Sealyx</title>
<meta name="description" content="Plans from $0. Free growth diagnosis, paid tracking and generated assets." />
<link rel="canonical" href="https://yoursite.com/pricing" />
<meta property="og:url" content="https://yoursite.com/pricing" />

Writing pages that both search engines and assistants can use

The same page has to serve two readers: a crawler deciding what you are relevant for, and a model deciding whether to name you in a sentence. The format that satisfies both is more specific than most marketing copy.

Answer the question in the first hundred words. Both a search snippet and an AI answer are extracted from the top of the page; if your first screen is a slogan and a hero image, there is nothing to extract. Lead with the direct answer, then explain.

Use concrete nouns and numbers. "Fast" is unusable; "under 200 ms to first byte" is quotable. "Affordable" is unusable; "$9 per month for three domains" is quotable. Facts survive the trip into someone else's answer; adjectives do not.

Structure with real headings. One H1 that states the topic, H2s that match the sub-questions people actually ask, and short paragraphs. Models retrieve passages, not pages, so a well-labelled section is what gets pulled.

Build the pages assistants reach for: comparisons, alternatives, pricing, and specific use cases. These carry high buying intent, are easy to write from what you already know, and are exactly what a shortlist question retrieves.

Comparison pages: "X vs Y" pages are cited constantly in AI answers because they contain both names and a stated difference. Be fair about the competitor; models penalise pages that read as pure marketing.
Alternatives pages: "Alternatives to X" captures buyers already in market. List real options including yours, with one honest line each.
Use-case pages: One page per job to be done, using the words the customer would use, not your internal feature names.
Pricing with numbers: A visible price is one of the strongest quotable facts you own. Pricing behind a contact form removes you from most recommendation answers.
Update dates: Show when a page was last reviewed. Freshness affects both crawl frequency and how confidently a model will cite it.

Questions people ask before signing up

Short answers to the things that decide whether Sealyx is right for you.

Do I need to install anything?: No. The free diagnosis needs a URL. Nothing is added to your site and no tracking script is required.
Does it work before launch?: Yes — connect a repository and audit the code. Attach the domain later and the history carries over.
Is my private code safe?: Authorisation is scoped, tokens are encrypted, and files are read only for the audit.
Which AI assistants are checked?: ChatGPT and Gemini today. Perplexity and Claude are on the roadmap.
Do I have to know SEO?: No. Every finding is written in plain English with the reason it matters and the exact change to make.
What does the free plan include?: The full growth diagnosis, AI visibility results and the technical SEO summary. Saved history, tracking and generated assets are on paid plans.
What is the difference between SEO and GEO?: SEO decides whether your page can be found and ranked in a list of links. GEO decides whether an assistant names your product inside a written answer. SEO rewards crawlability, relevance and links; GEO rewards retrievable text, a clearly stated category and quotable facts. A page can rank on Google and still never be mentioned by ChatGPT, and the reverse happens too.
Why does my site score badly if it looks fine in a browser?: Almost always because the content arrives after JavaScript runs. Your browser executes that JavaScript; most crawlers and AI agents read the first HTML response only. If that response is an empty shell, the page is effectively blank to them, no matter how good it looks to you.
Should I block AI crawlers to protect my content?: Blocking GPTBot or Google-Extended removes you from the answers those assistants generate, including the ones where a buyer is asking exactly what your product does. Blocking is a reasonable choice for paywalled archives; for a product site it usually costs distribution and returns nothing.
Does llms.txt actually do anything?: No engine is contractually bound to read it, and it will not rank you. What it does is give any agent that fetches your domain one short, unambiguous statement of what you sell, who it is for and what it costs — which is exactly the material an answer engine needs to quote. It takes ten minutes and cannot hurt.
How long until I see a change?: Technical fixes such as a missing sitemap, a noindex tag or a broken canonical are usually reflected within days of the next crawl. Ranking and AI-mention changes follow content and repeat crawling, so plan in weeks — the value of a scan is that it tells you the change landed, not that it happened instantly.
How often should I re-check?: After every deploy that touches routing, metadata or rendering, plus a scheduled weekly check. Most regressions we see are accidental: a noindex left over from staging, a canonical pointing at the homepage, or a framework upgrade that moves content behind hydration.
Do AI answers change between runs?: Yes, assistants are non-deterministic. That is why Sealyx sends several prompts per scan and reports a mention rate rather than a single yes or no, and why comparing two scans over time is more meaningful than reading one in isolation.

Glossary

The vocabulary used across reports.

SEO: Search engine optimisation — making a page findable and understandable to search crawlers.
GEO: Generative engine optimisation — making a product likely to be named inside AI-generated answers.
llms.txt: A plain-text file describing your product for language models, similar in spirit to robots.txt.
SSR: Server-side rendering — HTML delivered fully formed, which crawlers and AI agents can read without executing JavaScript.
JSON-LD: Structured data embedded in a page so engines know what it describes.
Bottleneck: The one dimension whose score is limiting growth the most right now.

Start with the free diagnosis

One URL, no signup, about forty seconds. You will see whether AI assistants name you, what is technically broken, and the single action worth doing next.

Audit my site