Auditing a B2B Page for AI Answer Engines: FITq, RACE and a BLUF Rewrite, Run on Our Own Page

A buyer who asks an answer engine which go-to-market motion to lead with, or which routing tool suits a regional courier, gets a short answer built from passages the engine could fetch and lift. A page that opens with a slogan, keeps its date in code and answers the question three scrolls down gives the engine nothing to lift, however good the product behind it.

The fixes that matter for answer engines are small and structural: the buyer's question as the heading, the answer in the first 100 words, a visible date and author, tables instead of styled boxes, and a plain line on who the product is not for.

This guide runs the AI Search Visibility (AEO/GEO) plugin, the partner plugin we built with Optise, on one of our own pages and shows every score it gave. If an audit is worth running, it should find problems on ours too, and it did.

Each step decides what the next one may change. The table alongside shows the question each tool answers.

FITq goes first because it starts with a scripted fetch of the page: HTTP status, time to first byte, page weight, headings, canonical tag, structured data, tables and lists, and whether the content needs JavaScript to appear. If the fetch fails, the skill will not score the page; if the page is a JavaScript shell, it caps Findability and will not score the rest. There is no point writing copy that crawlers will not see.

RACE goes second and reuses the same fetch, but asks a different question: could an AI agent doing a vendor evaluation pull out what you need to use the product, the steps, who it is not for and the proof? The skill warns that a page "can be FITq-A but RACE-F" or the other way round, so running both stops you fixing one signal and breaking another.

The BLUF Writer goes third, only for the gap the audits named: three answer blocks of about 40, 50 and 60 words in one of six fixed patterns, from proof points you give it. CRAFT Content goes last as a second reader, checking buzzwords, claims that need proof, fragments, merge fields and a missing ask.

Read the totals as ranges. The fetch is a script, but the scoring is Claude applying the skill's guide, and each audit skill holds two guides: descriptive bands for each signal, and point items that add up. On this page they disagree. The bands give 62 (grade C) and 49 (grade D). A strict count of the point items on the same fetch gives about 40 to 50 on FITq, a D or an F, and 2 to 4 on RACE, an F. I lead with the bands because they credit what the page says in prose, such as the implied constraints, which the point items score as zero. Because this is our own page, report both, and work from the reasons in each row.

Step 1. Score the page with the FITq audit

Input: the URL and the buyer question I wanted the page to win: "Which go-to-market motion should a B2B company lead with?" No EU market context, and the default reader (a marketing or growth lead).

What came back: 62 of 100, grade C, a band its scoring guide describes as "Fixable in 1-2 weeks". Findability took full marks: HTTP 200 with a time to first byte of 1,374 ms, 16.2 KB, a canonical tag, no JavaScript needed, and a robots.txt that allows GPTBot, ClaudeBot, PerplexityBot and the other AI crawlers by name.

How to read it: the page is easy to fetch and hard to quote. The H1, "Choose the GTM motion that should lead", is an instruction rather than the buyer's question, and the opening 23 words describe what EPIC does instead of saying which motion leads. Trust scored 10 because the page shows no date, no byline and no source for the "tends to lead" rules, and the date matters most: it caps Trust at 15 and is worth 8 points, against 5 each for a byline and sources. The date, 2 October 2026, sits only in the JSON-LD. The motion cards and example scores sit in styled boxes, not lists or a table, so there was little an engine could lift whole.

Its top five fixes, in order: a question-form H1 with an answer in the first 100 words, the date printed on the page, a named byline, an FAQ block with FAQPage markup, and the example scores as an HTML table. For a page selling into Europe, Trust this low would also have sent it to EU Trust Centre, in the same plugin.

Hand-off: the first fix needs new words, but I ran RACE before writing any.

Step 2. Check agent readiness with the RACE audit

Input: the same URL and the same fetch. I named no competitors, so every alternative it suggests stays a placeholder.

What came back: 49 of 100, grade D: "Major gaps. Page reads as marketing, not evaluation material." The inputs are on the page only in prose, there are no numbered steps, and the one worked example proves what the tool returns, not that the rules are right. Constraints was the best signal, because each motion card says when that motion "tends to lead" and the example admits "It is a first pass".

How to read it: an agent that reaches the page cannot tell what you need to run EPIC, how long it takes or when not to use it. Its first fix is a section titled "When EPIC is not the right first step", with starter text for three constraints and a [User to add] placeholder in place of every alternative. The skill's rule is never to publish that template "as if it were the user's actual constraints": ours are ours to write.

Hand-off: both audits wanted an answer at the top of the page, so one answer block went to the BLUF Writer.

Step 3. Write the answer block with the BLUF Writer

Input: the buyer question and five proof points copied from the page itself: the four motions scored from 1 to 10, the inputs (stage, deal size, sales cycle, market), and the rules for when Ecosystem and ABM, product-led growth, and inbound and outbound tend to lead. No voice sample, so a neutral tone. I used rewrite mode, because the page already has an opener.

What came back: a diagnosis of the current opener first: 23 words, "17 under the 40-word floor", trying to define a category but defining only the tool, with "four" as its only number. It chose Pattern 1, Direct Answer, because the question has a single answer that depends on conditions, and said why it passed over the Top 3 pattern: four motions that each win under different conditions do not rank. It wrote variants of 41, 50 and 57 words and recommended the 50-word one, in the table alongside.

How to read it: check every word against the page, because the skill checks its own rules, not your facts. Its pick holds two paraphrases, not one: "accounts are few" for "the market has few accounts", which it flagged, and "sales cycles" for the page's "deal cycles", which it did not. With no outcome to cite, it left a placeholder instead of a proof sentence: the gap RACE scored under Evidence.

My judgement on its first sentence: do not ship it. "A B2B company should lead with the motion that fits its stage, deal size, sales cycle and market" is close to circular: true and quotable, but it tells the buyer nothing they did not know. I would lead with concrete rules, chosen from the page's own words: two of its four Ecosystem conditions and the full product-led rule. That bends Pattern 1, whose first sentence should answer the whole question, because it names only two of the four motions; I accept that for a sentence a buyer can act on. I sent a 54-word version, rules joined by a semicolon, to CRAFT Content; its split produced the 57-word third row alongside.

Step 4. Give the passage a second check with CRAFT Content

Input: the new question-form H1, my 54-word answer block from Step 3, the other two motions' rules copied word for word from the page, and the call to action. I pasted them into the CRAFT Content improver as a landing page, with the audience (B2B founders and CMOs choosing a motion), the goal (score their own motions in GTM Alpha) and your_sector set to "other".

What came back: it marked the semicolon sentence "long sentence, 39 words", so I split it in two and ran it again (120 words in all). The second run scored 8.5 of 10: clarity and goal alignment 10, structure and engagement 8 each, for "Few transition words" and "Limited 'you' language". The Flesch reading score was 73, and it said "No automatic edits were found".

How to read it: a clean result means the wording passed, not the facts. The tool says so itself: "These checks do not judge whether your claims are true." The facts were checked in Step 3, against the page. I would take the "you" note for the call to action and leave the answer block in the third person, because an engine quotes it to someone who is not on our page, and I would set aside the transition-word note because each sentence of an answer block should stand on its own. Both are my own working rules, not the tool's. Its checklist also left "Social proof included" unticked: the RACE Evidence gap again.

Set your_sector: left out, it read my first passage as SaaS and added notes for software buyers.

Step 5. Merge the fixes into one list

The two audits gave ten fixes between them, with two overlaps: the opening answer block, which both wanted, and RACE's SoftwareApplication schema, which it merges with FITq's FAQPage markup in one JSON-LD block. The six in the table alongside are my selection for a first pass, not the skills'. I left off the SoftwareApplication schema and the "H2s as questions" half of FITq's last fix for now. The efforts are the skills' own labels, "ship this week" or "ship this month".

How to read it: the first two items need almost no new writing, since the answer block exists and the date and author are already in our structured data. The constraints section and the Evidence fix need thought, because only we can say when EPIC is the wrong first step, and only we can link each rule to a source or a worked case we can attribute.

When the fixes ship, run the FITq audit again. With the earlier score in Claude's memory, it flags any swing over 20 points and fetches again before reporting.

1. Pick one decision page and write down, in one sentence, the buyer question it must answer.
2. Run the FITq audit on it. If Findability is below full marks, fix that first.
3. Run the RACE audit on the same URL and merge the two fix lists into one.
4. Copy three to five proof points from the page, or from sources you can link, and run the BLUF Writer in rewrite mode.
5. Paste the new heading, the answer block and the call to action into the CRAFT Content improver with your sector set, and change only what it flags.
6. Ship the visible date and the byline the same day.

1. Rewriting before auditing. Copy written before the fetch can end up on a page engines cannot read.
2. Writing the answer block about the product instead of the question. Our old opener described EPIC, while the buyer asked which motion to lead with.
3. Keeping the date and author in code only. Our date sat in the JSON-LD and still cost Trust points.
4. Publishing the starter constraints. RACE's "not ideal for" text is a template full of placeholders; the constraints have to be yours.
5. Reading a high score as a citation. Both audits score page structure, and neither tells you whether an engine quotes you today.

• How to Turn One Verified Customer Result Into a Case Study, a LinkedIn Post and a Newsletter in the Founder's Own Voice
• From the Buyer Who Champions You to a Message Sales Will Use: Positioning With IMPACT, CRAFT Content and Revenue Enablement
• Agencies Sell Hours, We Ship GTM Tools: How to Chain the 81 Helix Tools Into One Quarter of GTM Work
• All 15 guides