Two Months Running an AI-API Reference Site Solo: Google Punished the Launch, Bing Didn't Care
I run ai-info — an independent reference site for AI API decisions: pricing breakdowns, model-vs-model comparisons, and guides, aimed at developers who search things like "cheapest coding agent plan" or "is X worth it." No affiliate links (frozen on purpose), nothing for sale in this post. Everything below is a snapshot as of October 8–9, 2026; index states and rankings drift daily, so treat the dates as part of the data.
What shipped
Two months, 42 published pieces (independent slugs on the sitemap as of Oct 8), six sections covering DeepSeek, model comparisons, coding agents, local models, and news. The first batch of 19 posts went out in one burst — deliberately, to give crawlers something to chew on.
The Google arc, in three acts
Act one, the honeymoon. The site got indexed fast — exactly one page: the DeepSeek hub. Not the homepage, not any article. One navigation page. It even earned a brief stretch of impressions. Encouraging, in a narrow way.
Act two, the fade. Sometime around September 18, impressions went to zero and the hub dropped out of the index. No manual action, no message in Search Console, just silence.
Act three, the present. As of the October 8 GSC reading: 3 pages indexed — homepage, news hub, local hub. Zero article pages. Forty-two pieces of content, and Google trusts exactly the skeleton.
Whatever the right label is (scaled-content suspicion, new-domain distrust, or just the post-hangover every search doc warns about), the practical read is the same: this is a site-level verdict, not a per-article one.
What I decided not to do: panic-slow the output to "fix quality." With Bing and direct traffic independently viable and 40+ pieces sitting as a starting library, the play is to keep publishing on a daily floor and point the backlink push at the skeleton pages — homepage and hubs — rather than deep articles. Hubs are what Google trusted once; that's the best signal I have.
The Bing surprise
On September 29 I cross-posted a cost-math piece to dev.to with the canonical pointing back to my original. Bing indexed the copy the same day and counted the backlinks. Google, as expected, never indexed the copy at all — which is exactly how canonical folding is supposed to behave, so no complaint there. It just underlines that the two engines are playing very different games with the same site right now.
That post carried 7 dofollow links back to the site, re-crawled on October 8 and still dofollow. So the channel works as a link source. It does not work as a ranking shortcut — Google never indexed the copy, so none of its on-page text helped my pages. Expect reach ≈ zero, expect credit = the links themselves. Budget your effort accordingly.
The workflow that makes 42 posts survivable
Volume only works if per-post quality has teeth. Three rules doing the heavy lifting:
- Narrow verification. Every post gets 3–5 hard numbers checked directly against official sources. Background figures cite the outlet they came from, by name. Verification stamps only go on numbers actually reconciled.
- The Chinese-media translation gap. I monitor Chinese tech outlets and vendor announcements. A DeepSeek pricing change announced in Chinese typically gets real English coverage a day or two later. Writing for that 6–48 hour window is the closest thing to an unfair advantage a solo operator gets.
- Dated caveats everywhere. Every post carries an "as of" statement. I'm only accountable for correctness at publish time — prices and plans drift, and pretending otherwise is how a reference site loses trust.
Honest scoreboard
Two months in: Google 0 (3 skeleton pages, no impressions), Bing 1 (same-day indexing, backlinks counted), direct and referral traffic small but real and growing. The bet underneath all of it: decision-intent queries around AI APIs are still under-served in English, and backlinks plus time still compound for genuinely useful pages.
Check back in six months. Or don't — I'll be posting the numbers either way.
Snapshot caveats: index counts from GSC on Oct 8–9, 2026; sitemap slug count from a live crawl on Oct 8, 2026; dev.to link attributes re-verified Oct 8, 2026. All of this changes — verify anything you'd cite.
Related from the site: DeepSeek API hub · V4 API cost calculator · Why the same DeepSeek call costs 2x more by the hour · Context caching, the one discount everyone has
Top comments (0)