What SmoothSeen checks, and what it does with what it finds
From the Professional plan, every analysis checks whether ChatGPT and Gemini actually cite your site and measures it again every month or week. Alongside that citation score come two more, from 0 to 100: Web ranking measures what Google needs to crawl, understand and show a page, and AI ranking measures whether an AI assistant can pull an answer out of it and trust it as a source. And it tells you what to fix first and which of your competitors already does it.
- Scores
- Web and AI, from 0 to 100; plus AI citations, from Professional
- Web ranking
- 8 categories
- AI ranking
- 20 categories in 5 groups
- Plans
- The same checks on all of them; you start with 7 days of the Agency plan
On this page
Web ranking
What a search engine needs to reach the page, understand it and show it properly. The checks are grouped into these categories:
- Security and trust
- HTTPS, security headers and email authentication. A browser warning costs more visits than any optimisation gains.
- Crawling and indexing
- Whether search engines can reach the page, understand it, and tell which version is the canonical one.
- Content
- What Google reads to work out what the page is about: title, description, headings and enough text.
- Speed
- How long the page takes to respond, render and react. LCP, INP and CLS are the Core Web Vitals Google uses for ranking; TTFB, FCP and TBT help explain why they fail.
- Technical optimisation
- Page weight: images, CSS and JavaScript that can be trimmed without changing anything the visitor sees.
- Structured data and brand
- The markup that describes the page to search engines and social networks —rich results (stars, prices, breadcrumbs), the preview when it is shared— and whether the search engine recognises the brand: its declared profiles and whether it shows up when its name is searched.
- Mobile and accessibility
- Google indexes the mobile version first. This also covers barriers that stop people using the page with a keyboard or screen reader.
- Links and authority
- How many sites link to the domain, the state of its own links, and whether traffic is being measured.
AI ranking
These are two different questions. AEO: can an answer be taken from the page without reading all of it? GEO: why would a model choose it as a source when writing its own answer? The categories are grouped into five blocks.
AI access
Whether AI assistants' search crawlers can get in, whether the page lets them quote it, and whether their agents can operate it.
- AI engine access
- Which AI crawlers robots.txt lets in. Blocking the search ones (OAI-SearchBot, Claude-SearchBot, PerplexityBot…) can take you out of their answers; blocking the training ones (GPTBot, ClaudeBot…) does not take you out of ChatGPT, Claude or Perplexity, and it is a legitimate choice.
- Navigable by AI agents
- Whether an AI agent (ChatGPT, Claude, Gemini) can understand and operate the page: buttons, links and fields with names, and valid ARIA. It is what the agent reads when it books, buys or fills in a form for the user.
Extractability (AEO)
Whether the text arrives in the HTML the server sends, whether the answer sits where an extractor looks for it, and whether it comes in a format it can copy as it is.
- Text in the served HTML
- Whether the text arrives in the HTML the server sends. ChatGPT, Claude and Perplexity do not run JavaScript: whatever the browser paints afterwards does not exist for them.
- Direct answer
- Whether the page answers the question in its opening paragraphs. An assistant quotes what it can extract without reading the whole page.
- Structure for AI
- Headings and semantic HTML that let a model tell which part of the page answers what.
- Extractable format
- Summaries, tables of contents, FAQs, lists and a question-and-answer tone: the blocks an assistant can quote as they are, plus the llms.txt that tells it what to read.
- Readability
- Sentences and vocabulary a model — and a person in a hurry — can process without rereading.
Credibility (GEO)
Why a model would trust the page as a source: data, references, authorship, date and markup.
- Citable facts
- Concrete figures, dates and prices. A model cites what it can verify before it cites generic claims.
- Sources and citations
- Links to external sources and attributed quotes: what separates documented content from opinion.
- Domain authority
- The site’s accumulated reputation, measured by who links to it.
- Freshness
- A visible published or reviewed date. Without one, an assistant cannot tell whether the fact still holds.
- Authorship and experience
- Who signs it and with what credentials. The E-E-A-T signal that weighs most on sensitive topics.
- Data consistency
- That what the page says matches what directories and external listings say.
- AI markup
- Structured data (Schema) that explicitly declares what the content is about.
Coverage
Whether there is enough content and it resolves what people are looking for.
- Topic coverage
- Whether the page genuinely resolves the search intent or merely mentions it.
Specific to the page type
Each page is classified as a product page, a catalogue, a local business, an article or undeclared, and is only asked for what belongs to its type: a product page is not asked for an author byline.
- Product page
- Price, availability, ratings and a description of its own: what an assistant needs to recommend a purchase.
- Local business
- Address, opening hours, coordinates and a Google Business profile, to show up in searches with local intent.
- Catalogue
- Browsable categories, shipping and returns on the shop front page.
- Service or application
- Prices or plans, ratings and contact details of a service or an application: what an assistant needs to recommend it.
- Page type
- What this page is: a product page, a catalogue, a local business or an article. If it never says, an assistant cannot place it and has no idea which questions it would answer.
AI citations
From the Professional plan
The checks tell you whether an assistant can cite you; this section tells you whether it does. In every analysis, SmoothSeen writes the questions a customer who doesn’t know the brand yet would ask, puts them to ChatGPT and Gemini with web search turned on, and gives a 0–100 citation score, separate from the web and AI ranking scores. You don’t have to write any list of questions, though you can swap them for your own.
- 5 questions per analysis on Professional and 10 on Agency (also in the free trial), in your account’s language.
- For each question and each engine: whether the answer links to your site and in what position, only names your brand, or leaves you out.
- “Cited instead of you”: the domains the assistant cites when it doesn’t cite you, and how many answers cite the competitors you chose.
- Those domains are also suggested as competitors and go through the same checks as your site: that is how you see why they get cited and what to fix first.
- It is a snapshot: assistants don’t always answer the same way. That is why the test has its own PDF, “AI citations”, and runs again in every tracking analysis, so you can see how the citation score evolves.
Local business
A page counts as a local business if it declares LocalBusiness markup (or one of its subtypes) or has a Google Business Profile confirmed on its own domain. The local checks are then added: what an assistant needs to answer “where is it?” or “is it open now?”.1
- Local business
- Address, opening hours, coordinates and a Google Business profile, to show up in searches with local intent.
- Data consistency
- That what the page says matches what directories and external listings say.
1The Google profile is taken into account when the website it declares is on the same domain as the page analysed.
Online shops
A shop has two kinds of page and each is measured on its own terms. The product page, on what an assistant needs to recommend the purchase. The home page or catalogue, on what it needs to answer “where do I buy this?”.
- Product page
- Price, availability, ratings and a description of its own: what an assistant needs to recommend a purchase.
- Catalogue
- Browsable categories, shipping and returns on the shop front page.
Action plan
The findings from both scores, turned into a task list ordered from most to least serious: where to start, and how to argue for it in front of the client.
- Ordered by what matters most to fix, grouped by area.
- Each task says what to do and why it matters, in language you can show your client.
- If you have measured competitors, which of them already meet each task.
- Separately, what isn't a task but changes how search engines and assistants see you.
Competitor comparison
From the Professional plan
Up to 3 competitors per client, measured with the same checks as your site. The result is not a table of scores: it is what they meet and you do not, ordered by how many of them meet it.2
- They are suggested from who appears on Google for the queries the analysis simulates for your page and from the domains ChatGPT and Gemini cite instead of you. You can change them by hand.
- Your competitors go through the same checks as your site, so the comparison is fair.
- Your advantages also appear, and so does what none of them meet: the chance to be the only one that has it.
- Each competitor counts as one of the month’s analyses, every time you re-measure it too, and takes up no space in your client list.
2Gaps are ordered first by how many competitors already pass the check.
Before and after
You run the analysis again whenever you want, or tracking does it for you, and see what changed between two analyses of the same client: what was fixed, what is new and what is still pending. It answers “what did you do this month?” better than a score that goes up a few points, and it comes out as a PDF progress report for your client: their scores, how they moved, what changed and the next steps.
- You can compare with any of the 12 previous analyses of the same client; by default, the one immediately before.
- Both analyses are compared with the same yardstick: what changes is your site, not the ruler.
- You see what was fixed, what is new and what is still pending.
PDF reports and a share link
From the Professional plan
The report leaves the dashboard in two ways, with the same data you see on screen, from the Professional plan. On Professional they carry the SmoothSeen brand; on Agency, yours.
- Six documents: web ranking, AI ranking, AI citations, competitor comparison, action plan and progress report.
- Link
- An address your client opens without an account, with the whole report —search and AI visibility, AI citations, keywords, competitors, tracking and the action plan— and its PDFs. Search engines do not index it, you can see how many times it has been opened, and you can revoke it: it stops working for good.
White label
On the Agency and Enterprise plans
Deliver reports under your agency’s brand instead of ours.
- Name, logo, main colour, website and contact email. Any field you leave empty uses SmoothSeen’s.
- It applies to the PDFs, to the share link (with the brand of the agency that owns the report, not of whoever opens it) and to the tracking alert email.
- You can switch it off whenever you like: reports go back to carrying the SmoothSeen brand.
Tracking and alerts
From the Professional plan
Each client is re-analysed on its own, without you having to remember: monthly on Professional, and monthly or weekly, as you choose, on Agency and Enterprise. With every analysis its history grows, and you can see where it is heading.
- You choose which page is watched. If you choose none, it is the last one you analysed for that client.
- Each scheduled analysis is a full analysis, with its citation test, and counts as one of the month’s analyses; if there are none left that month, it resumes when they renew.
- The client’s page shows how its three scores, web, AI and citations, evolve, and marks which analyses were automatic and which you launched by hand. Before and after compares two of them.
- If a score drops clearly against the previous analysis of that page, you get an email with the drop and a link to the report.3
3The alert needs a clear drop, not the small variations that are normal between one measurement and the next. The citation score needs a wider margin, because assistants do not always answer the same way.