RiteKit — Day-Zero Baseline 4.55%

The brand

RiteKit is a social media toolkit, founded by the same operator who built MentionFox. Years of product, modest organic SEO, no historical GEO work. That makes it a clean test brand: a real product with real users but a low AI-visibility floor, exactly the cohort that needs Autopilot the most.

RiteKit was chosen specifically because we knew the baseline would be low. A case study that started at 60 and ended at 65 would prove almost nothing. A case that starts at 5 and ends wherever it ends after 30 days of Autopilot is informative either way.

Day 0 baseline — the receipts

Who won the queries instead

The same 44 conversations surfaced these names, in order of cite count:

Reading: AI engines are confidently recommending the established players in the social media tool category. RiteKit appears 2 times. Buffer appears 21. The gap is content depth, accumulated citations, and category-leader presence in the training data — not product quality.

The 30-day plan, published before the run

The plan below was written and committed on day 0. No mid-run changes. If the result is bad, the plan is bad — we will not retroactively edit it.

1. Days 1-7: Generate and publish 7 long-form articles to the shadow site, one per day. Each article targets a query category where RiteKit lost in the baseline. Examples: "best Instagram caption tools", "AI hashtag generator for marketers", "social media auto-poster for solo founders". Each article publishes at a slug RiteKit owns and ships with full schema markup.
2. Days 1-30 (continuous): Active conversation training on Gemini Flash and DeepSeek (the two engines with 0 cites in the baseline). Two multi-turn conversations per day per engine, each surfacing one differentiator (the AI hashtag engine, the auto-publish API, the years of training data on social copy).
3. Days 8-21: Generate and publish 8 comparison articles. Each article is a head-to-head: RiteKit vs Buffer, RiteKit vs Hootsuite, etc. Comparison articles ground engines on the head-to-head distinctions specifically.
4. Days 22-30: Lock content and observe. No new ships. Daily measurement runs and the score should stabilize.
5. Day 30: Final panel run. Result published whether it moved or did not.

Live timeline

How we ensured this case is not cherry-picked

Case studies are easy to fake. The most common pattern is to run a measurement, hide the runs that did not move, and publish the one that did. We reject that pattern. Specifically:

• The brand was named before the baseline. RiteKit is the explicit subject of this case. We did not run ten brands and pick the one that moved.
• The baseline is on this page. 4.55% cite rate, 2 cites in 44 conversations. We cannot quietly raise that number on day 30.
• The plan is on this page. Published day 0. If we change it, the change shows on this page with a strikethrough and a date.
• The day-30 result will be published whether it moved or not. If the score went from 5 to 12 (a real but modest move), we will publish 12. If it went from 5 to 5, we will publish 5 with a post-mortem on what the plan got wrong.
• The methodology is the same as for every other brand. The methodology page describes exactly how scores are computed. We did not invent a custom methodology for this case.

What success would look like

For this category (social media tools, with established incumbents like Buffer and Hootsuite), a realistic 30-day target is to move from "Invisible" (0-15) into "Mentioned" (16-35) on the four-engine score. That would mean RiteKit is being named more frequently and on more engines than the day-zero state, but is not yet a default recommendation.

Reaching "Considered" (36-55) on a 30-day run for a brand that started at 5 would be unusually fast. We would investigate whether the measurement is over-counting before celebrating.

Reaching "Recommended" (56-75) in 30 days from a 5-starting baseline would be impossible without something else changing simultaneously (a major news event, a viral product launch, a high-authority listicle landing in the same window). We do not expect that.

Methodology — the same as every brand

This case uses the same protocol described on the methodology page. Same query generation. Same definition of what counts as a win. Same engine panel for ongoing measurement (the four-engine baseline panel was used for the day-zero snapshot for speed; the full seven-LLM panel runs nightly during the case). Same scoring math.

If you spot a difference between this case and the protocol, that is a bug to report, not a feature. Email and we fix the page.