Tabellab

Your First AI Visibility Audit: a Prompt Set, Mention Rate, Citation Rate and the Rate by Intent

Lesson 1 of 20Lab

All lessons
  1. Your First AI Visibility Audit: a Prompt Set, Mention Rate, Citation Rate and the Rate by IntentLabFree
  2. Weighting the Prompt Set: Demand Weights, the Weighted Mention Rate with SUMPRODUCT and the Gap to the Simple AverageLab Locked
  3. Mentions vs Citations: Citation Share, Linked-Mention Share, the Overall Share from Totals and a ChartLab Locked
  4. Share of Voice vs Competitors: Total Mentions, Share of Voice, Rank and the Gap to the LeaderLab Locked
  5. Position in the Answer: a Weighted Score with SUMPRODUCT, Prominence per Mention and the Weighted Share of VoiceLab Locked
  6. One Prompt Set, Four Assistants: Runs, Mention Rate and Citation Rate per Assistant with SUMIFS and the Strongest AssistantLab Locked
  7. Referral Traffic from AI Assistants: a Channel Mapping with XLOOKUP, Sessions per Channel, Conversion Rate and the AI ShareLab Locked
  8. Which Pages Earn from AI Referrals: Engagement Rate, Conversion Rate, Revenue per Session, an Index with ROUND and a ChartLab Locked
  9. Content Gap Scoring: Gap to the Best Competitor, a Score from Demand and Business Value, Rank, Priority Labels and a ChartLab Locked
  10. Where the Answers Come From: Cited Domains, Citation Share by Source Type, Uncovered Third-Party Pages and an Outreach RuleLab Locked
  11. Tracking over Time: Weekly Mention Rate, a 4-Week Rate, the Trend with SLOPE, a FORECAST.LINEAR Projection and a Line ChartLab Locked
  12. Noise or Signal: Standard Error, Margin of Error, the Lower Bound and the Runs Needed for a Reliable Mention RateLab Locked
  13. Did the Content Update Work? Before and After Rates, a Control Group and the Difference-in-DifferencesLab Locked
  14. Tone and Truth: Sentiment Counts, Net Sentiment, Error Rate per Assistant with SUMIFS and a Rule that Flags Wrong AnswersLab Locked
  15. The Cluster × Assistant Grid: Mention Rates with Two-Criteria SUMIFS and Mixed References, the Spread per Cluster and a Heat MapLab Locked
  16. Head-to-Head with a Competitor: Exclusive and Shared Answers, Presence Rate, Co-Mention Share and Prompts Where You LeadLab Locked
  17. Forecasting AI Referrals: Month-over-Month Growth, the Compound Monthly Rate, a Linear and a Compound Projection and a Line ChartLab Locked
  18. What-If Grid: Citation Rate × Click Rate, Monthly Revenue with Mixed References, the Citation Rate Needed and the Target ZoneLab Locked
  19. Prioritising the Backlog: Extra Mentions per Hour, Rank, Cumulative Hours in Rank Order and a Do-or-Later DecisionLab Locked
  20. Final Project: AI Visibility Scorecard by Assistant with Mention Rate, Share of Voice, Citation Share, Referral Sessions, Status Flags and a Combo ChartLab Locked

You are not signed in — the lab works, but your progress is not saved. Sign up

Lesson 1 of 20 is free. Unlock all 20 hands-on labs, the final exam and your certificate.

Buy course — $9.90
Loading the spreadsheet…

Why AI answers need their own measurement

For twenty years, being found online meant ranking on a results page. You typed a query, got ten blue links, and the position of your link decided how many people clicked. That model is no longer the only one. A growing share of product research now starts with a question typed or spoken to an AI assistant, or ends in an AI-written summary above the classic results. The reader gets a finished paragraph: a short explanation, three or four brand names, sometimes a handful of source links. There is no position seven. A brand is either part of the answer or it is not.

This changes what a marketing team can measure. Rank trackers, impression counts and click-through rates describe a list of links. An answer has none of these. What it does have are two observable events: the answer names your brand, or it does not; the answer links to your site as a source, or it does not. Count these events over a well-chosen set of questions and you get the two basic measures of visibility in AI search: the mention rate and the citation rate. The wider discipline is sometimes called answer engine optimisation or generative engine optimisation.

This free lesson shows how to build that first audit in a spreadsheet. The example is Quillfern Outdoors, a fictional online retailer of hiking and camping gear with the made-up domain quillfern.example. The company, its competitors, the prompts and every number on this page are invented for learning, and any benchmark is illustrative. The table below is the same data you work with in the lab above, so you can check each formula against the text.

What a prompt set is, and what makes a good one

You cannot ask an assistant "how visible are we?". You can only ask it what a customer would ask and look at the answer. A prompt set is therefore the measuring instrument of AI visibility: a fixed list of realistic customer questions that you send to an assistant again and again, always in the same wording. It plays the role a keyword list plays in classic search tracking, with one difference. Prompts are longer and more conversational than keywords, because people talk to assistants in full sentences and add their situation: wide feet, a budget, a trip in wet weather.

  • Start from real questions. Good sources are your on-site search log, support tickets, sales chats, the questions in review sections and the long-tail queries in your search console export. Avoid inventing prompts that flatter the brand.
  • Cover the buying journey. Include informational prompts (how to choose, how to fix), comparison prompts (best, versus, under a budget) and transactional prompts (where to buy, which shop). Each intent behaves differently, as the example will show.
  • Keep brand prompts apart. A prompt that contains your own name, such as a versus question, will mention you almost every time. It is useful for checking what is said about you, but it inflates the mention rate. Mark it, or report it separately.
  • Freeze the wording. Small changes in phrasing change the answer. A series is only comparable from month to month if the prompts stay identical. Add new prompts as a new version of the set and note the date.
  • Stay small enough to maintain. Twelve prompts are enough to learn the method. A working set for a mid-sized shop often has 50 to 150 prompts grouped into topics. Quality and stability matter more than size.
Assistants personalise and localise. The same prompt can give different answers depending on country, language, account history and the model version in use. Run audits in fresh sessions without history, from the market you sell in, and write down the date and the assistant. A number without this context cannot be compared with next month's number.

Mentions, citations and the two basic rates

Because assistants generate text with a random element, the same prompt does not return the same answer twice. Ask once and the brand may be there; ask again and it is gone. A single try is an anecdote. The audit therefore sends every prompt several times, each time in a new session, and treats each answer as one run. Quillfern uses 20 runs for the prompts it cares most about and 10 for the rest.

Every run is then coded by a written rule. It counts as a mention when the answer text names Quillfern as a brand or shop, whether in praise, in a neutral list or in criticism (tone is a separate code that a later lesson adds). It counts as a citation when the answer shows a source link to quillfern.example. A run can be a mention without a citation: the assistant recommends the brand but links to a review site. Less often it is a citation without a mention: the assistant uses your buying guide as a source and never names you in the text. The two rates are simple divisions:

  • Mention rate = runs with a brand mention ÷ runs
  • Citation rate = runs with a citation to your site ÷ runs

From prompt set to mention rate and citation ratePrompt set12 fixed promptsRuns180 answersCodingone rule for all66 mentionsmention rate 36.7%32 citationscitation rate 17.8%Same wordingevery month10 or 20 runs each,fresh sessionsMention = brand namedCitation = link to our siteRate = coded runs ÷ all runs: 66 ÷ 180 and 32 ÷ 180
The audit pipeline at Quillfern Outdoors: 12 prompts, 180 runs, one coding rule, two rates. All numbers are fictional.

Why track both? The mention is about reputation: does the assistant consider the brand worth naming? The citation is about evidence and traffic: are your own pages good enough to be used as a source, and does the reader get a link to click? A brand that is mentioned but never cited is known through other people's pages. A brand that is cited but rarely mentioned has useful content and a weak commercial profile. The remedies differ, so the measures must stay separate.

A fully worked example: Quillfern Outdoors, September 2026

In September 2026 the Quillfern team ran its 12 prompts on one assistant, 180 runs in total, and coded every answer. The first five columns of the table are the raw audit; the last two are the rates you calculate in the lab.

PromptIntentRunsMentionsCitationsMention rateCitation rate
best waterproof hiking boots for wide feetComparison209445.0%20.0%
how to break in leather hiking bootsInformational102220.0%20.0%
lightweight 2-person tent under $300Transactional2011655.0%30.0%
Quillfern vs Northpeak rain jacketsComparison10105100.0%50.0%
how to choose a sleeping bag temperature ratingInformational101010.0%0.0%
where to buy merino base layers onlineTransactional207335.0%15.0%
best trekking poles for beginnersComparison206230.0%10.0%
how to waterproof a tent seamInformational10000.0%0.0%
best 40 litre backpack for weekend hikesComparison2012760.0%35.0%
is down or synthetic insulation better in wet weatherInformational103130.0%10.0%
hiking boot resole service by mailTransactional10000.0%0.0%
best camping stove for two peopleComparison205225.0%10.0%
All prompts180663236.7%17.8%

The two calculated columns need one formula each, written for row 4 and copied down to row 15:

  • Mention rate (F4): =D4/C4. Runs with a brand mention divided by runs. For the wide-fit boot prompt that is 9 ÷ 20 = 45.0%.
  • Citation rate (G4): =E4/C4. Runs with a citation divided by runs. For the same prompt, 4 ÷ 20 = 20.0%.
  • Format both columns as percentages. A rate above 100% is impossible; if you see one, numerator and denominator are swapped.

Walk through a few rows. The backpack prompt is Quillfern's best neutral prompt: 12 of 20 answers name the brand (60.0%) and 7 link to the site (35.0%). The tent prompt follows with 11 of 20 (55.0%). The versus prompt reaches 100.0%, which says nothing about visibility, because the prompt itself contains the brand name; its interesting number is the citation rate of 50.0%, meaning half the answers took facts from Quillfern's own pages. At the other end, two prompts never produced a mention in ten runs each: "how to waterproof a tent seam" and "hiking boot resole service by mail". The sleeping bag prompt shows the mention-without-citation case: one mention, no link.

Mention rate per prompt, 12-prompt setInformationalComparisonTransactionalOverall 36.7%Waterproof boots, wide feet45% (9/20)Break in leather boots20% (2/10)2-person tent under $30055% (11/20)Quillfern vs Northpeak jackets100% (10/10)Sleeping bag rating10% (1/10)Buy merino base layers35% (7/20)Trekking poles, beginners30% (6/20)Waterproof a tent seam0% (0/10)40 litre backpack60% (12/20)Down or synthetic when wet30% (3/10)Boot resole by mail0% (0/10)Camping stove for two25% (5/20)0%25%50%75%100%
Mention rate of each prompt, coloured by intent. The dashed line is the overall mention rate of 36.7%.

The overall mention rate: divide totals, do not average rates

One number for the whole prompt set is useful for a monthly report. The correct calculation divides the totals: all mentions by all runs. Quillfern has 66 mentions in 180 runs, so the overall mention rate is 36.7%. In the lab that is =SUM(D4:D15)/SUM(C4:C15) in cell F16. The overall citation rate is built the same way: 32 ÷ 180 = 17.8%.

The tempting shortcut is =AVERAGE(F4:F15), the average of the twelve prompt rates. It returns 34.2%, which is a different number. The reason is that the prompts do not have the same number of runs. Averaging the rates gives a prompt with 10 runs the same weight as a prompt with 20 runs, although it contributed only half as many observations. Here the six 20-run prompts happen to have higher rates than most 10-run prompts, so the simple average comes out too low. In another month it could be too high. The rule holds everywhere in this course: add up the counts first, then divide.

Whenever you report a rate, keep the two counts next to it: "36.7% (66 of 180 runs)". The reader can see how much data is behind the number, and you can re-aggregate later by intent, topic or month without going back to the raw answers.

Mention rate by intent with SUMIFS

An overall rate hides structure. Quillfern tagged each prompt with the buyer's intent, and the three intents behave very differently. To calculate a rate per intent you need conditional sums: the mentions of all prompts with a given intent, divided by the runs of those prompts. SUMIFS(sum_range, criteria_range, criterion) does exactly this. The formula in J4, copied down to J6, is =SUMIFS($D$4:$D$15,$B$4:$B$15,I4)/SUMIFS($C$4:$C$15,$B$4:$B$15,I4). The dollar signs lock the data ranges; only the reference to the intent label in column I moves down.

IntentPromptsRunsMentionsMention rateCitation rateFormula in J
Informational440615.0%7.5%=SUMIFS($D$4:$D$15,$B$4:$B$15,I4)/SUMIFS($C$4:$C$15,$B$4:$B$15,I4)
Comparison5904246.7%22.2%=SUMIFS($D$4:$D$15,$B$4:$B$15,I5)/SUMIFS($C$4:$C$15,$B$4:$B$15,I5)
Transactional3501836.0%18.0%=SUMIFS($D$4:$D$15,$B$4:$B$15,I6)/SUMIFS($C$4:$C$15,$B$4:$B$15,I6)

Mention rate and citation rate by intentMention rateCitation rate0%10%20%30%40%50%15.0%7.5%Informational6 of 40 runs46.7%22.2%Comparison42 of 90 runs36.0%18.0%Transactional18 of 50 runs
Mention rate and citation rate of the Quillfern prompt set by search intent (fictional data).

The picture is typical. Informational prompts (how to break in boots, how to seal a seam) are answered with instructions. Shops are rarely named, and Quillfern appears in only 6 of 40 runs, 15.0%. Comparison prompts ask for recommendations, so several brands are listed and Quillfern reaches 46.7%, helped by the versus prompt. Transactional prompts name sellers; Quillfern is at 36.0%, with a clear hole at the resole service, which the shop offers but apparently nobody on the web describes.

Notice that the denominator carries the same condition as the numerator. A frequent error is to divide the mentions of one intent by the runs of all prompts. For the comparison prompts that would give 42 ÷ 180 = 23.3% instead of 46.7%, and the three results would add up to the overall rate instead of describing three separate groups.

Step by step in the spreadsheet

  1. Lay out the audit with one row per prompt: the prompt text, the intent, the number of runs, the runs with a mention and the runs with a citation. Keep counts as plain numbers, not as text such as "9 of 20".
  2. In F4 type =D4/C4 and press Enter. Select F4 and drag the fill handle down to F15, or double-click it. Apply the percent format.
  3. In G4 type =E4/C4 and copy it down to G15 in the same way.
  4. In F16 type =SUM(D4:D15)/SUM(C4:C15) for the overall mention rate. Resist the average of column F.
  5. List the three intents in I4:I6. In J4 type =SUMIFS($D$4:$D$15,$B$4:$B$15,I4)/SUMIFS($C$4:$C$15,$B$4:$B$15,I4). Press F4 while the cursor is in a range to add the dollar signs. Copy down to J6.
  6. Check the result: the runs of the three intents (40, 90 and 50) must add up to 180, and no rate may exceed 100%.

How to read the result

An overall mention rate of 36.7% means that a little more than one answer in three names Quillfern on the questions the team chose. Whether that is good depends entirely on the prompt set and on the competitors, which is why later lessons add a demand weight per topic and the share of voice against rival brands. What the first audit already delivers is a map of strengths and holes.

  • High mention rate, high citation rate (backpacks, tents): the assistant knows the brand and uses its pages. Protect these topics: keep the pages current and specific.
  • Decent mention rate, low citation rate (trekking poles, camping stoves): the brand is named from third-party knowledge. Your own pages on the topic are probably thin or hard to quote.
  • Zero mentions (seam sealing, resole service): either no suitable content exists, or the assistant does not connect the brand with the topic. These are candidates for the content gap list.
  • Brand prompts (the versus question): ignore the mention rate, read the answers for accuracy and tone, and watch the citation rate.
Ten or twenty runs per prompt are a small sample. A prompt that shows 2 mentions in 10 runs could easily show 0 or 4 next time without anything having changed. Treat single-prompt rates as hints and base decisions on pooled numbers: the whole set, an intent, a topic cluster. A later lesson of the course calculates the margin of error and the number of runs you need.

Common mistakes and how to spot them

  • One run per prompt. The result flips between 0% and 100% from one audit to the next. Use at least ten runs, more for the prompts that matter.
  • Averaging percentages. =AVERAGE(F4:F15) gives 34.2% instead of 36.7%. Divide total mentions by total runs.
  • Swapped division. =C4/D4 returns values above 1, displayed as several hundred percent, and a division error where the mentions are zero.
  • Mixing mentions and citations. Dividing citations by mentions (E ÷ D) is a meaningful ratio, but it is not the citation rate. The citation rate divides by runs.
  • Unlocked ranges in SUMIFS. Without dollar signs the ranges slide down when the formula is copied, and the second and third intent sum the wrong rows.
  • Changing the prompts between audits. A new wording is a new prompt. Compare like with like, and version the prompt set.
  • Reading mentions as traffic. A mention sends no visitor by itself. Visits are measured on your own site as referral sessions, a separate lesson.

Collecting the data in practice

The arithmetic is the easy part; disciplined collection is what makes the numbers trustworthy. You can run a small audit by hand: open a fresh session, paste the prompt, save the answer, repeat. For larger sets teams use tracking tools or the providers' programming interfaces, keeping in mind that an interface without web access may answer differently from the consumer app that real buyers use. Whatever the method, a few habits apply.

  • Store the raw answers with date, assistant, model name if shown, prompt version and run number. You will want to re-code them later for position, sentiment or accuracy.
  • Write the coding rule down, including edge cases: is a product name without the shop name a mention? Does a link to a marketplace listing count as a citation? (In this course: a citation is a link to your own domain.)
  • Have a second person code a sample of answers. If the two of you disagree often, the rule is not clear enough yet.
  • Respect the providers' terms of use and rate limits, and do not try to manipulate answers with hidden text, fake reviews or invented expert quotes. The durable way into an answer is accurate, specific, well-structured content and a real reputation on independent sites.

Quick reference: the measures of this lesson

MeasureDefinitionFormula in the labQuillfern result
Mention rate per promptRuns with a mention ÷ runs=D4/C40% to 100%
Citation rate per promptRuns with a citation ÷ runs=E4/C40% to 50%
Overall mention rateAll mentions ÷ all runs=SUM(D4:D15)/SUM(C4:C15)36.7%
Overall citation rateAll citations ÷ all runs=SUM(E4:E15)/SUM(C4:C15)17.8%
Mention rate by intentMentions of the intent ÷ runs of the intent=SUMIFS(…D…)/SUMIFS(…C…)15.0% / 46.7% / 36.0%

What you will build in the rest of this course

Lesson 1 is free. Lessons 2 to 20 form the full course; each has its own live lab with a new Quillfern data set and builds on the audit you have just made.

Weighting and sharing the answer (lessons 2 to 5). You give each prompt topic a demand weight and calculate a weighted mention rate with SUMPRODUCT, separate citation share from mention rate, compute share of voice against seven competitors with rank and gap to the leader, and weight mentions by their position in the answer.

Assistants and traffic (lessons 6 to 8). You split a tracking log by assistant with SUMIFS, map referring domains to an AI channel with XLOOKUP, measure the AI share of sessions, and judge landing pages by engagement, conversion and revenue per session.

Finding and choosing the work (lessons 9, 10 and 19). You score content gaps by demand and business value, analyse which third-party domains the assistants cite and where your brand is missing, and rank a backlog of actions by expected mentions per hour until the available hours are used.

Tracking and testing (lessons 11 to 13). You follow the mention rate over twelve weeks with a pooled 4-week rate, SLOPE and FORECAST.LINEAR, put a margin of error on every rate, and test a content update against control topics with a difference-in-differences.

Quality and competition (lessons 14 to 16). You audit sentiment and factual accuracy of the answers, build a cluster × assistant heat map with mixed references and conditional formatting, and run a head-to-head against one rival with shared and exclusive answers.

Planning and reporting (lessons 17, 18 and 20). You project AI referral sessions with a linear and a compound model, explore a what-if grid from citation rate to revenue, and finish with a scorecard per assistant that combines mention rate, share of voice, citation share, sessions, status flags and a combo chart.

Summary

Visibility in AI search starts with a fixed prompt set and repeated runs. Each answer is coded for a brand mention and for a citation of your own site. The mention rate per prompt is =D4/C4, the citation rate =E4/C4. The overall rate divides the totals, =SUM(D4:D15)/SUM(C4:C15), and at Quillfern Outdoors that is 36.7% for mentions and 17.8% for citations. SUMIFS in numerator and denominator splits the rate by intent: 15.0% informational, 46.7% comparison, 36.0% transactional. Keep the prompts, the coding rule and the run counts stable, and the same sheet becomes the first point of a time series.

The lab at the top of this page holds the same twelve prompts. Building the audit there takes roughly a quarter of an hour, each formula is checked as you enter it, and the finished sheet can serve as a template for your own first prompt set.

Frequently asked questions

What is a mention rate in AI search?

The mention rate is the share of AI answers that name your brand: runs with a brand mention divided by all runs of a prompt or a prompt set. In a spreadsheet it is a simple division such as =D4/C4.

What is the difference between a mention and a citation?

A mention is the brand name appearing in the answer text. A citation is a source link to your own website shown with the answer. An answer can mention a brand without linking to it, and it can cite a page without naming the brand.

How many times should I run each prompt?

Answers vary from run to run, so one run is not reliable. Ten runs per prompt are a practical minimum for learning the method, twenty or more for important prompts. Decisions should rest on pooled numbers across many prompts, because single-prompt rates from ten runs have a wide margin of error.

How do I calculate an overall mention rate for a prompt set?

Divide the total number of mentions by the total number of runs, for example =SUM(D4:D15)/SUM(C4:C15). Do not average the individual prompt rates, because prompts with fewer runs would then count as much as prompts with more runs.

What is a prompt set?

A prompt set is a fixed list of realistic customer questions that you send to an AI assistant repeatedly in the same wording. It is the measuring instrument of AI visibility, comparable to a keyword list in classic search tracking, and should cover informational, comparison and transactional questions.

Does a high mention rate mean more website traffic?

Not directly. A mention without a link sends no visitor. Traffic depends on citations that readers click and is measured on your own site as referral sessions from the assistants' domains. Mention rate, citation rate and referral traffic are three separate measures.

Keep going: AI Search Visibility & Answer Engines: Prompt Sets, Mention & Citation Share, Share of Voice, AI Referral Traffic & Content Gap Scoring

This free lab is lesson 1 of 20. Unlock the other 19 hands-on labs, the final exam and your certificate for a one-time $9.90.

  1. Weighting the Prompt Set: Demand Weights, the Weighted Mention Rate with SUMPRODUCT and the Gap to the Simple Average
  2. Mentions vs Citations: Citation Share, Linked-Mention Share, the Overall Share from Totals and a Chart
  3. Share of Voice vs Competitors: Total Mentions, Share of Voice, Rank and the Gap to the Leader
  4. Position in the Answer: a Weighted Score with SUMPRODUCT, Prominence per Mention and the Weighted Share of Voice
  5. One Prompt Set, Four Assistants: Runs, Mention Rate and Citation Rate per Assistant with SUMIFS and the Strongest Assistant
  6. Referral Traffic from AI Assistants: a Channel Mapping with XLOOKUP, Sessions per Channel, Conversion Rate and the AI Share
  7. Which Pages Earn from AI Referrals: Engagement Rate, Conversion Rate, Revenue per Session, an Index with ROUND and a Chart
  8. Content Gap Scoring: Gap to the Best Competitor, a Score from Demand and Business Value, Rank, Priority Labels and a Chart
  9. Where the Answers Come From: Cited Domains, Citation Share by Source Type, Uncovered Third-Party Pages and an Outreach Rule
  10. Tracking over Time: Weekly Mention Rate, a 4-Week Rate, the Trend with SLOPE, a FORECAST.LINEAR Projection and a Line Chart
  11. Noise or Signal: Standard Error, Margin of Error, the Lower Bound and the Runs Needed for a Reliable Mention Rate
  12. Did the Content Update Work? Before and After Rates, a Control Group and the Difference-in-Differences
  13. Tone and Truth: Sentiment Counts, Net Sentiment, Error Rate per Assistant with SUMIFS and a Rule that Flags Wrong Answers
  14. The Cluster × Assistant Grid: Mention Rates with Two-Criteria SUMIFS and Mixed References, the Spread per Cluster and a Heat Map
  15. Head-to-Head with a Competitor: Exclusive and Shared Answers, Presence Rate, Co-Mention Share and Prompts Where You Lead
  16. Forecasting AI Referrals: Month-over-Month Growth, the Compound Monthly Rate, a Linear and a Compound Projection and a Line Chart
  17. What-If Grid: Citation Rate × Click Rate, Monthly Revenue with Mixed References, the Citation Rate Needed and the Target Zone
  18. Prioritising the Backlog: Extra Mentions per Hour, Rank, Cumulative Hours in Rank Order and a Do-or-Later Decision
  19. Final Project: AI Visibility Scorecard by Assistant with Mention Rate, Share of Voice, Citation Share, Referral Sessions, Status Flags and a Combo Chart

Buy course — $9.90 See the full course