Hypertune AEO

Bobby's AEO plan, revised 15 Oct: review

Thursday 15 October 2026, 15:45 Dubai

For Som, from Vin. Please read this alongside the AEO campaign report, and we'll discuss both.

What this is. A review of Bobby's revised AEO plan (7-page PDF, last changed Thu 15 Oct, 09:24 Dubai), checked against the six things we asked for on 13 Oct, our Ahrefs account and the work already on the tracker.

Bobby's plan: https://drive.google.com/drive/u/0/folders/1uxq02hcFxic28sl_WtjR-LN0PRKaBcNt

SEO tracker: https://hypertune-reports.pages.dev/seo-gantt-2026-10-02

Bottom line

The plan is good enough to discuss on a call, but not to approve yet. The method is right, and Bobby answered most of the 13 Oct asks: owners, hours, dates, tool and reporting are all in, and he built on what already exists instead of rebuilding it.

Two things the approval depends on are still missing: the price (he left it for the call) and a target (he pushed it to Fri 13 Nov). The timeline is also slower than it needs to be: the first baseline report lands on 13 Nov and the first AEO page around 20 Nov.

Recommendation: ask Bobby to send his price before the call. On the call, approve Phases 1 and 2 (setup and baseline) only, with the baseline brought forward to the Fri 30 Oct bi-weekly report. Decide Phases 3 and 4 once the baseline shows the gaps, as Bobby himself proposes.

Quality of the plan

Overall: 6 out of 10. A big step up from the 13 Oct scope, which had three of its four phases undefined, 30 to 60 questions to track by hand, no owners, hours or dates, and rebuilt things we already have. This version is clear, well organised and honest about what's new. It falls short on the commercial side (no price, no target) and on pace.

AreaScoreWhy
Structure and clarity8/10Seven clear sections, each saying what's new versus what already exists. Easy to follow
Building on existing work8/10Reuses the facts list, the AI Assistant channel, the survey and the bi-weekly report. Folds the AEO check into his SEO check
Method7/10Measure first, then fix what AI tools miss or get wrong, ranked by the sites they cite. Sound and standard. Thin on what actually moves AI answers: Reddit, YouTube and creator videos, roundups and Wikidata are mentioned but not planned
Evidence5/10Uses real figures (Google Ads terms, GA4 landing pages), but two headline numbers have no clear source and the "5 questions tracked today" aren't in any tool we can see
Owners5/10Every task has owners, but three tasks have three to five each, and Kamil is listed as an owner
Realism of time4/10Two weeks for about 4 hours of setup. Review hours for Austin and Vinny are high. Content hours may double-count the existing calendar
Commercial2/10No price and no target, the two things the approval depends on, both asked for on 13 Oct

What this means: the plan works as a menu of the right activities, but it isn't something we can approve or run as it stands. A fuller campaign, built on this plan and the SEO work already running, is in a separate report.

What the plan is, in short

AEO (answer engine optimization) means getting Hypertune named, cited and described correctly in answers from ChatGPT, Perplexity, Gemini, Copilot and Google's AI results.

Bobby's plan has seven parts:

  1. Measurement. Grow the tracked questions from 5 to 20 to 30, run them on each AI tool with Ahrefs Brand Radar, and log whether Hypertune is named, how it's described, which competitors appear and which sites are cited.
  2. AI crawler access. A one-off check that AI search crawlers can read the site.
  3. Content built for AI answers. Find the gaps in what AI tools say, map the follow-up questions, write self-contained paragraphs, and publish original data reports (the first one from Hypertune's per-game FPS numbers). Added to the AI content framework, not a new process.
  4. Fact control. Check AI answers against the approved facts in the AI content framework. No separate claims register.
  5. Off-site sources. Rank directories, roundups, Reddit and Quora by which ones the AI tools actually cite, and add captions to Hypertune's YouTube videos.
  6. Monitoring and correction. Trace wrong answers to their source and fix them there.
  7. Reporting. AI referral visits, Bing's Copilot citations and Search Console, reported in the bi-weekly report.

The six asks from 13 Oct

AskAnswered?What the plan says
1. PriceNoSetup and baseline are a one-off payment; after that, a monthly fee by the hour. No rate or total. Bobby left it for the call
2. HoursYesBy person and phase. Bobby: 15 hours one-off, then 16 hours a month
3. OwnersMostlyEvery task has owners, matching our default split. Three tasks have three or four owners each
4. DatesYes, but slowFour phases from 15 Oct, baseline report Fri 13 Nov, first AEO page about Fri 20 Nov
5. Tool and sizeYesAhrefs Brand Radar, $50 a month for 2,500 checks or $100 for 7,000 (we don't have it yet; costs below). Starts at 20 to 30 questions, as asked
6. Q4 targetNo"Named on more tracked questions than at the baseline, and described correctly." Rough targets set with Vinny on Fri 13 Nov

Also asked on 13 Oct, and now in the plan:

Hours by phase

PersonPhase 1: setupPhase 2: baselinePhase 3, per monthPhase 4, per month
Bobby4111616
Som2688
Vinny0.5166
Austin1.5166
Kamil (through Austin)up to 3noneup to 5up to 3
Team total11194139

Bobby's own time is 15 hours one-off (Phases 1 and 2), then 16 hours a month. Phase 3 runs about five weeks (16 Nov to 18 Dec).

Cost: unknown until Bobby gives his hourly rate. Brand Radar comes on top: about $50 a month (see "What Brand Radar costs").

Timeline

PhaseBobby's datesWhat happens
1. Approval and setupThu 15 to Fri 30 OctQuestion set built (5 to 20 to 30) and approved by Vinny. Brand Radar loaded. Analytics, Bing, Search Console and server log figures pulled
2. BaselineMon 2 to Fri 13 NovBaseline runs done Fri 6 Nov. Crawler check and Som's competitor research. Baseline in the bi-weekly report Fri 13 Nov, with the Phase 3 target
3. Content, fixes, trackingMon 16 Nov to Fri 18 DecFirst brief, first page to Vinny and Austin about Fri 20 Nov, then 2 guides a week from 23 Nov. Monthly re-test from 7 Dec. Check against the baseline Fri 18 Dec
4. OngoingFrom Mon 21 DecMonth to month: tracking, the bi-weekly report, guides, more questions and placements if the Phase 3 check supports it

Issues

  1. No price. Nothing can be approved without it. Ask for it before the call, not on it.
  2. No target. A rough Q4 number is possible now, even before the baseline. For example: Hypertune named on X of the 25 questions, and no wrong statements on price, the trial or the Intel and AMD wording.
  3. Slower than it needs to be. Two weeks for setup covers about 4 hours of Bobby's time: building 25 questions and loading them into a tool that runs the checks automatically. The baseline could be in the Fri 30 Oct bi-weekly report instead of 13 Nov. That would bring the first AEO page into early November, and give two or three readings before Christmas instead of one.
  4. It may double-count content. "About 9 guides" and "2 guides a week" sound like the existing content calendar, which already runs two pages a week. If they're the same pages, the AEO work changes how they're written, not how many there are, and the content hours shouldn't be counted twice.
  5. The writer is out of date. The plan names Som as the only writer. Since the 15 Oct call, Hoa also writes: the gaming guide and every game page from Valorant onwards.
  6. The review hours look high. It gives Austin and Vinny 6 hours a month each. Austin's part could be signing off claims and data reports, about 1 hour a month.
  7. Shared owners. YouTube captions has three owners (Vinny, Austin, Kamil), error tracing has four (Bobby, Som, Austin, Kamil), and content creation has five. Each task needs one owner.
  8. Kamil is listed as an owner. On the SEO plan, website changes are Austin's steps, sent through Linear. They should read "Austin (Kamil, through Linear)".
  9. Crawler data from the wrong place. It asks Austin for server logs. Cloudflare already shows AI crawler visits, so no logs are needed. The access check itself was done on 13 Oct: the crawlers behind ChatGPT search, Claude search and Perplexity can read the site. Only the training crawlers are blocked (GPTBot, ClaudeBot, CCBot), which is Austin's choice.
  10. Two baseline numbers need a source.
  11. "17.6K Google AI impressions" from a Search Console "Generative AI report". Which report is this, and what does it count?
  12. "167 Copilot citations, all to the homepage", with the time window still to confirm.
  13. Check volume decides the package. See "What Brand Radar costs" below. Most setups fit the $50 package; daily tracking or many phrasings per question would need the $100 one.

What Brand Radar costs

We don't have Brand Radar today. Its question tracking ("custom prompts") is priced by checks: one check is one question on one AI tool, each time it runs. Prices from Ahrefs' pricing and help pages, 15 Oct:

OptionCost a monthWhat you get
Included in our Lite plan$0150 checks a month (5 questions a day on one AI tool). ChatGPT, Perplexity, Gemini, Copilot, Google AI Overviews and Google AI Mode
Basic package$50+2,500 checks, then $0.02 per extra check
Growth package$100+7,000 checks, then $0.015 per extra check
Scale package$250+25,000 checks, then $0.01 per extra check
Full Brand Radar AI, one AI tool$199Research any brand across Ahrefs' database of AI answers
Full Brand Radar AI, all tools$699The same for every tool, plus 2,500 checks

The packages are month to month, can be cancelled at any time, and Ahrefs doesn't refund used months. Claude costs 8 checks per run instead of 1.

What Bobby's plan would use, at 25 questions on 6 AI tools:

How oftenChecks a monthPackage
Once a month150Free on Lite
Once a month, each run 3 times (Bobby's plan)450$50
Every two weeks, each run 3 times900$50
Every two weeks, 3 runs and 3 phrasings each2,700$50 plus about $4 extra, or $100
Every day4,500$100

Recommendation: the $50 package covers the plan, as Bobby says. The full Brand Radar AI ($199 to $699) is for researching other brands across Ahrefs' whole database and isn't needed. The first baseline round (25 questions on 6 tools, once) fits in the 150 free checks we already have, so the package only needs buying once regular tracking starts.

What I checked

Questions for Bobby

Price, before the call

  1. What's your hourly rate, and the one-off total for setup and baseline (15 hours)?
  2. What's the monthly cost after that (16 hours)? Is Brand Radar ($50 or $100) on top?

Timing

  1. Can setup finish by Fri 23 Oct, with the baseline in the Fri 30 Oct bi-weekly report rather than 13 Nov?
  2. Can the re-test run every two weeks, matching the reports, so there are two or three readings before Christmas?

Scope

  1. Are the "9 guides" and "2 a week" the existing content calendar pages, or extra ones?
  2. You're taking over monthly question tracking, so do we drop it from Som's November list? Do llms.txt and the claims page stay with Som?
  3. What's the turnaround for your AEO check on a page: 24 or 48 hours?
Target
  1. Can you give a rough Q4 target now, to refine after the baseline?
  2. Which 5 questions are tracked today, and where?

Tool and data

  1. Can the first baseline round run on the 150 free checks in our Lite plan, with the $50 package bought only once regular tracking starts?
  2. Which AI tools will you track, and how many phrasings per question?
  3. Where do the 17.6K Google AI impressions come from, and what's the window for the 167 Copilot citations?
  4. Can crawler visits come from Cloudflare instead of server logs?

Owners

  1. One owner each for YouTube captions, error tracing and content creation.
  2. Can Austin's time come down to signing off claims and data reports, about 1 hour a month?

Next steps

  1. Bobby sends his price before the call.
  2. Book the call (on the tracker for Vin and Bobby since 15 Oct).
  3. On the call: approve Phases 1 and 2 at the agreed price, with the baseline in the Fri 30 Oct report and a rough Q4 target.
  4. After the call: the tracker gets the approved phases, dates and owners.