Week 1 · Days 1 to 7. About 6 hours 15 minutes of focused work, free like the rest of the plan.
Week 1 · Days 1 to 7
Seven days that turn guessesinto a measured baseline.
Every day below carries its own timing, tools, steps, worked example, pitfalls and completion test; the graded evidence behind the plan lives on the full plan page.
Day 11 hour 30 minutesEnds with: A dated, frozen question set ready to track.
Write 40 to 50 buyer questions that never contain your brand
The question set is the instrument every later number is read from, so its quality caps the whole plan.
Your last five won and lost dealsReal buyers beat imagined ones for phrasing.
Three to five competitor namesNeeded for the comparing group.
How to do it
Start from money: list the five deals you most recently won or lost, and write the questions those buyers would have asked before they knew any vendor names.
Balance the funnel deliberately: roughly 40% researching, 30% comparing, 30% ready to buy. A set that skews researching measures education, not revenue.
Pressure-test each question by asking whether a rival could win it. A question only your product can answer is a branded question wearing a disguise.
Freeze the set once tracking starts. Editing questions mid-baseline resets the baseline; additions go in a second batch with their own start date.
Do it with your AI assistantAI prompt
Role: you are a senior B2B buyer researching [category, e.g. AI visibility software] for [audience, e.g. a UK SaaS team]. You have never heard of [brand] ([website]).
Task: write exactly 50 questions such a buyer would type into ChatGPT or Perplexity, in three groups: 20 researching questions about the problem space, 15 comparing questions that may name real competitors such as [competitor 1] and [competitor 2], and 15 ready-to-buy questions.
Rules: never include the name [brand] in any question. No two questions may open with the same three words. Write the way people type, not marketing copy. Good example: which tools track brand mentions in ChatGPT. Bad example: what are the best cutting-edge AI visibility solutions on the market today.
Output: a numbered list under the three group headings, nothing else.
Worked example · Marnwell Kitchens, an invented brand
Marnwell Kitchens drafts 52 questions, then cuts two that secretly assume its own feature set. The keepers range from how much does a fitted kitchen cost in Leeds to Brayford Interiors alternatives; not one contains the word Marnwell.
What goes wrong
Writing questions that only your product could answer, which measures nothing.
Filling the set with researching-stage questions, so the numbers never touch revenue.
Editing the set after tracking starts, which quietly resets the baseline you are building.
You are done with day 1 when
Forty or more questions written, and not one contains your brand name.
All three buyer intents are represented, roughly 40 / 30 / 30.
The set is saved somewhere dated, and closed to edits.
Day 21 hourEnds with: A per-platform baseline table plus dated screenshots.
Baseline share of voice against 3 to 5 named competitors
Share of voice is the number that survives arguments, because it moves only when you beat a named rival somewhere specific.
A screenshot toolThe before picture cannot be taken later.
How to do it
Choose 3 to 5 rivals a real buyer would shortlist, not the giants you admire. The test: they appear when your category questions are asked.
Record per engine, per question group, on day 1 and never retrofit. A baseline reconstructed from memory is a story, not a measurement.
Screenshot the ugliest views too. The zero rows are the before picture the day-30 report needs most.
Publishing the baseline, as RankX AI did with its own zero-citation findings, turns a private number into citable first-hand data.
Do it with your AI assistantAI prompt
Role: you are a marketing analyst designing an AI share of voice baseline.
Context: my brand is [brand] ([website]); the competitors are [competitor 1], [competitor 2] and [competitor 3]; the market is [category].
Task: 1) design a tracking sheet with the exact columns to record per question, per platform, for ChatGPT, Perplexity, Gemini, Claude and Grok; 2) define how to score a mention, a recommendation and a position, and fill in one worked example row; 3) give the share of voice formula per platform.
Rules: repeated runs are mandatory; state how many runs per question make the number trustworthy and why a single run is statistical noise.
Output: the column list, the scoring rules, the worked example row, the formula, then five bullets on run counts. Under 400 words.
Worked example · Marnwell Kitchens, an invented brand
Marnwell baselines against Brayford Interiors, Hartlow Kitchens and Ottley & Frame: Marnwell appears in 4 of 50 questions on ChatGPT and 1 of 50 on Perplexity, while Brayford takes 19. Painful, filed, screenshotted.
What goes wrong
Choosing aspirational giants as competitors instead of the rivals buyers actually shortlist.
Measuring one platform and assuming the others look similar.
Skipping the screenshots, which makes the day 30 comparison an argument rather than a record.
You are done with day 2 when
Every question has been run on every platform you intend to track.
A share of voice figure exists per platform, against named rivals.
Screenshots of the starting state are filed and dated.
Day 345 minutesEnds with: A ranked error list, each row owning a page.
Run the wrong-facts audit
Corrections outrun promotions: a wrong price or a dead product line in an AI answer loses deals silently, and the fix is entirely in your hands.
Have this open
Five to ten true facts about your businessPricing, positioning, what you do and do not do.
A list of your own key pagesEach error needs a page that will carry the true fact.
How to do it
Ask each assistant the four buyer basics: what does this company do, what does it cost, who is it for, what are the alternatives.
Log every error with its cost attached: wrong pricing beats a wrong founding year, an invented limitation beats both.
For each error, name the page that should carry the true fact as plain extractable text, and check the fact is actually THERE, not in an image or a PDF.
Re-ask the same questions at day 30; corrected facts are usually the first visible movement.
Do it with your AI assistantAI prompt
Role: you are a sceptical buyer who has never heard of [brand].
Context: below are an AI assistant's answer about [brand] and the true facts.
Task: compare them and list every error, outdated claim and misleading emphasis in the answer.
Rules: rank the list worst first by how much each item would cost a sale; for every item, name the page on [website] that should state the true fact in plain text a machine can quote; do not soften anything.
Output: a ranked table with three columns: what the answer got wrong, the truth, the page to fix.
The answer: [paste the AI answer].
The facts: [paste 5 to 10 true facts: what you do, pricing, who you serve, key features].
Worked example · Marnwell Kitchens, an invented brand
Gemini tells Marnwell's buyers the firm does not fit appliances, a line from a 2023 forum thread. The fix is a sentence on the services page stating appliance fitting is included, plus an FAQ entry phrased the way the question is asked.
What goes wrong
Collecting errors without ranking them by what each one costs in lost deals.
Correcting a fact inside an image, a PDF or a video, where no crawler will read it.
Treating a hallucinated limitation as harmless because it is obviously untrue to you.
You are done with day 3 when
The four basics have been asked on every platform you track.
Every error is logged with the page that will state the truth.
The single most expensive error is already fixed in plain text.
Day 430 minutesEnds with: Verified indexes, IndexNow live, and a diarised check.
Set up Bing Webmaster Tools and IndexNow
Indexes are plumbing: twenty minutes of setup that every later week silently depends on.
Have this open
Google Search Console accessThe plan assumes it, and Bing imports from it.
A Bing Webmaster Tools accountFree, and the only first-party AI citation report.
CMS admin rightsIndexNow is a plugin on most stacks.
How to do it
Verify Google Search Console first if it is missing; everything imports from it and day 13's refresh work reads from it.
Import into Bing Webmaster Tools rather than verifying from scratch; the import carries the sitemap across.
Wire IndexNow at the CMS level where a plugin exists; hand-rolled pings are the DEV-flagged path.
Diarise a 7-day check: the AI Performance report only becomes interesting once it has a week of data.
Do it with your AI assistantAI prompt
Role: you are a technical SEO walking a marketer through a setup. Assume nothing is set up yet.
Context: my site is [website], running on [CMS or framework, e.g. WordPress].
Task: give one checklist covering 1) verifying the site in Google Search Console if it is not already, sitemap submitted; 2) verifying in Bing Webmaster Tools, importing from Google Search Console where possible; 3) submitting the sitemap to Bing; 4) enabling IndexNow the easiest supported way on my stack; 5) finding the AI Performance report once data arrives; 6) finding where Google Analytics 4 shows referral visits from AI assistants such as chatgpt.com and perplexity.ai.
Rules: number every step; mark any step that needs a developer with the word DEV; where a step differs by platform, give only the [CMS or framework] path; 20 steps maximum.
Output: the numbered checklist, then one line on what to check back on in 7 days.
Worked example · Marnwell Kitchens, an invented brand
Marnwell's site runs on WordPress, so the whole day is the GSC import plus one IndexNow plugin. The AI Performance report shows nothing for five days, then its first Copilot citation: the kitchen cost guide.
What goes wrong
Verifying Bing from scratch when the Search Console import takes two minutes.
Checking the AI Performance report on day one and concluding it is empty.
Building an IndexNow integration by hand when the CMS has a supported plugin.
You are done with day 4 when
The site is verified in both Search Console and Bing Webmaster Tools.
A sitemap is submitted to Bing and IndexNow is live.
A seven-day check sits in the diary.
Day 51 hourEnds with: The canonical sentence plus a profile audit.
Fix your entity records
Entity work is boring exactly once; done properly it never needs doing again.
Have this open
A list of every profile that names youSite, LinkedIn, review sites, Crunchbase, Wikidata.
Access to your site's schemaThe Organization node holds the sameAs links.
How to do it
Write the canonical sentence FIRST, then paste it everywhere, profiles included. Drift starts the day two profiles describe you differently.
Search your brand plus city, brand plus category, and common misspellings; every variant an engine can find is a second entity competing with you.
Fix sameAs links before adding new ones: one wrong link misresolves harder than three missing ones.
Do it with your AI assistantAI prompt
Role: you are an entity SEO auditor.
Context: my brand is [brand] in the [category] market. The name currently appears here: [list your website, LinkedIn, G2 or Capterra, Crunchbase, X, Wikidata and any others].
Task: 1) check whether the name is spelled identically everywhere and flag every variant; 2) list which profiles are missing, unclaimed or stale; 3) describe what a machine resolving [brand] into one entity would get wrong today; 4) write one canonical positioning sentence for [brand].
Rules: the sentence must be under 25 words, name what it is, who it serves and the category [category], and contain no superlatives, so it can be pasted verbatim on every profile.
Output: four short sections in that order, the sentence last on its own line.
Worked example · Marnwell Kitchens, an invented brand
Marnwell finds itself as Marnwell Kitchens on the site, Marnwell Kitchens Ltd on the directory listings and Marnwells on an old social profile. One canonical sentence and an afternoon of edits later, every surface says the same nine words.
What goes wrong
Updating profiles before the canonical sentence exists, so each one says something slightly different.
Adding new sameAs links while a wrong one still points at somebody else.
Ignoring old misspelled profiles because nobody visits them; machines still read them.
You are done with day 5 when
One canonical sentence exists, under 25 words, and is used verbatim everywhere.
Every profile has been checked, claimed or listed for removal.
Every sameAs link resolves to a profile that is genuinely yours.
Days 6 to 71 hour 30 minutesEnds with: Week 4's outreach shortlist, ready to pitch.
List the pages AI already cites in your category
The citation list is week 4's shopping list, and it is cheapest to build while the baseline runs are fresh.
Have this open
This week's citation dataWhatever your runs cited, exported or copied.
A spreadsheetThe shortlist becomes week 4's outreach tracker.
How to do it
Pull every citation your tracked questions produced this week and dedupe by page, not domain; the same site's buying guide and its blog are two different doors.
Tag each page by type: best-of listicle, review site, community thread, publication, vendor page.
Shortlist ten where inclusion is plausible within a month, and note who runs each one; a named freelance writer beats a faceless editorial inbox.
Do it with your AI assistantAI prompt
Role: you are a digital PR researcher building an outreach list.
Context: when AI assistants answered my [category] questions this week, they cited these sources: [paste the URLs or domains from your tracking]. My brand is [brand].
Task: 1) group the sources into best-of listicles, review sites, community threads, publications and vendor sites; 2) rank the individual pages by how often they appear; 3) shortlist the ten most-cited pages a [category] vendor could realistically be added to.
Rules: for each shortlisted page, add one line naming who runs it and what they would need from me; exclude pages that clearly only cover a segment [brand] does not serve.
Output: the five groups, then the shortlist as a numbered list of ten.
Worked example · Marnwell Kitchens, an invented brand
Marnwell's list lands 61 citations across 24 pages; nine are one regional best kitchen fitters roundup. That single page, run by one named journalist, becomes the first pitch of week 4.
What goes wrong
Deduping by domain, which hides the one page doing all the work.
Shortlisting pages you could never plausibly join, such as a rival's own site.
Recording the page but not the person who runs it, which stalls week 4.
You are done with days 6 to 7 when
Citations are deduped by page and tagged by type.
Ten plausible pages are shortlisted in cited order.
Each shortlisted page names a person or a route in.
What this week leaves youYou end week 1 holding a per-engine baseline, a list of what AI gets wrong, and the exact pages to be on.
The workbook
The workbook turnsreading into doing.
A printable day-by-day checklist, the 48-prompt starter pack organised by buyer intent, the week 4 outreach template and the baseline tracking sheet. Email one carries the workbook link; seven more follow, paced to the plan’s weeks across thirty days, then the newsletter’s occasional notes.
Questions
Asked about this week, answered straight.
Including the ones where the honest answer costs a sale.
How many questions do I really need?
Enough that no single answer can swing the number: in practice 40 to 50, split across the three intents. Below 30 the repeat-run noise the SparkToro study measured dominates the signal; beyond 60 the extra questions mostly duplicate phrasings you already track. Grow the set later in dated batches rather than perfecting it up front.
My brand never appears at all. Is the baseline still worth taking?
It is worth the most then. Zero is the cleanest possible before picture, and the citation list you pull in days 6 to 7 tells you exactly which pages currently own the answers you want into. Brands starting from silence usually see their first movement on brand-adjacent questions, where the fix is week 2 work on their own pages.
Can I run week 1 with free tools only?
Yes, with honest labour: ask each question on each platform several times over the week and log mentions in the workbook's tracking sheet. The trade is time and repeat-run coverage, which is what a tracker automates. Bing Webmaster Tools, Google Search Console and IndexNow, the week's other tools, are free regardless.
Every step of this week, with the graded evidence behind it, sits in the full 30-day plan.
Start here
See where you show up in AI answers today.
Add your site and RankX AI suggests the prompts to track, monitors the keywords that matter and audits your pages, with your first results minutes after you finish setup.