Frase Review

For twenty years there was one question in content marketing: does this page rank. That question now has a sibling, and the two do not always have the same answer.

People search Google and they ask assistants. Google ranks pages; assistants name sources. A site engineered for the first can be entirely absent from the second, and you usually find out from a customer weeks later.

Frase's framing is that both are one job and should run through one system — which is either the correct read of where search is going or a considerable bet on it.

Frase: Ranking On Google And Cited By AI

• Both search behaviours in one loop — audit, research, write, optimise, publish, monitor.
• An autonomy dial: publish autonomously, keep every page under review, or anywhere between.
• AI visibility monitoring tracking which of your pages assistants actually cite.
• Runs from Claude, Cursor and any MCP tool, alongside a CLI and API.
• 70-plus languages written natively rather than translated, with publishing to WordPress, Webflow, Sanity and Wix.

What Is GEO And How Does It Differ From SEO?

Frase splits its optimisation into two: content optimisation that scores and lifts pages for Google, and GEO optimisation that tunes content to get cited by AI.

The distinction is real, and it comes down to what each system is doing with your page.

A search engine ranks documents. It decides which page best answers a query and sends the person there, so the work is winning a position and earning a click.

An assistant extracts an answer. It reads several sources, synthesises, and names some of them — so the work is being the source whose material is clearest to lift, and whose claim is specific enough to quote.

That changes what good looks like in a way most content teams have not internalised. Long preambles, keyword-padded introductions and answers buried nine paragraphs down survive in a ranking system and disappear in an extraction one. Frase's own audit output names exactly that failure — an answer buried in paragraph nine — which is a sharper diagnosis than a keyword report.

A search engine ranks documents and sends a click. An assistant extracts an answer and names a source. Those reward different pages.

Why AI Visibility Is Hard To Measure

Frase offers AI visibility monitoring, tracking which of your pages AI tools mention, plus free GEO score and AI visibility checkers.

Measuring this is genuinely difficult, and understanding why keeps the numbers in proportion.

Rankings are stable and public. The same query returns roughly the same results for everyone, so a position can be checked and tracked. AI answers are none of those things — they vary by phrasing, by user, by session, by model version, and often between two identical requests.

So AI visibility is a sampling exercise rather than a measurement. Someone asks a set of questions repeatedly across several assistants and counts how often you appear, which produces a directional signal rather than a position.

That is still worth having. Going from never mentioned to sometimes mentioned is a real change, and the trend is meaningful even when the absolute number is soft. Just do not treat it as a rank, and ask any vendor which assistants they sample, how often, and with what queries.

The Autonomy Dial And Publishing On Its Own

Frase describes an autonomy setting explicitly: let it publish on its own, keep every page under review, or anywhere in between — you set the dial.

That framing is a better piece of product thinking than the automation itself.

Content automation triggers an entirely reasonable fear. A tool that publishes to your live site without review can put something wrong, off-brand or embarrassing in front of customers, and the damage is public before anyone notices.

Most vendors respond by underselling the automation. Frase instead makes the level an explicit, adjustable decision, which lets a team start at full review, watch what the system would have published, and move the dial as trust accumulates.

That is the right adoption path, and it is worth using deliberately. Start at review, keep it there for a month of real output, and only then decide — because the question is not whether the tool can publish autonomously but whether you would have approved what it produced. Whether any of it converts is a different measurement again, and Cometly's job rather than a content platform's.

• Start at full review and keep it there for a month.
• Ask which assistants a visibility number samples, and how often.
• Run the free checkers on your own site before buying anything.
• Fix buried answers before writing anything new.
• Treat AI visibility as a trend, never as a ranking position.

Audit, Fix Pack And Content Guard

The entry point is an audit that scores every page for search and for AI answers, orders the most valuable fixes first, and packages them — its example shows a readiness score, three prioritised problems and a fix pack of forty-four items.

Leading with an audit rather than a subscription is smart positioning, because it inverts the usual sale.

Most content tools ask you to commit and then discover value. An audit shows you what is currently costing you before you change anything, which is a far easier internal conversation than a licence request.

The three example problems are also well chosen, because each names a distinct failure: a guide AI never cites, a page whose ranking slipped from three to seven, and an answer buried in paragraph nine. Those are different diseases with different treatments.

Content Guard is the other half — catching ranking decay and shipping the fix. Decay is the most under-managed problem in content marketing precisely because nothing announces it; a page just quietly stops working, and without monitoring nobody looks until a quarterly report. Product pages decay the same way, at a scale only a bulk tool like Describely can address.

Running Frase From Claude Or Cursor

Alongside the interface, Frase offers an MCP server, a CLI and an API, described as running it from Claude, Cursor and any MCP tool.

For a marketing product that is an unusual thing to build, and it says something about where the company thinks the work is going.

The conventional assumption is that a marketer opens a content platform and uses it. An MCP server assumes something else: that the person is already working inside an assistant, and the platform should be reachable from there rather than requiring a context switch.

For technical content teams that is genuinely useful — briefs, audits and optimisation checks become things you ask for in the middle of writing rather than tasks you leave to do.

It also hedges the platform's own risk. If the interface people work in keeps moving, being available through a standard protocol matters more than owning the destination — the same reasoning that puts Komo and other search products in front of agents rather than only in front of people.

Running A Content Programme For Both Searches

The temptation with a platform this broad is to start producing. That is backwards — the fastest returns are almost always in pages you already have.

This sequence starts where the value is highest and moves the automation dial slowly.

1. Run the audit before writing a single new page.
2. Fix buried answers on pages that already rank — that is the cheapest GEO win.
3. Define brand voice properly before letting anything draft.
4. Keep autonomy at full review for a month and read what it would have shipped.
5. Watch AI visibility as a trend across several assistants, not one.
6. Let Content Guard find decay rather than auditing manually each quarter.

Frase: Modules And Languages